Object Detection Using YOLO v3 Deep Learning

This example shows how to train a YOLO v3 object detector.

Deep learning is a powerful machine learning technique that you can use to train robust object detectors. Several techniques for object detection exist, including Faster R-CNN, you only look once (YOLO) v2, and single shot detector (SSD). This example shows how to train a YOLO v3 object detector. YOLO v3 improves upon YOLO v2 by adding detection at multiple scales to help detect smaller objects. Moreover, the loss function used for training is separated into mean squared error for bounding box regression and binary cross-entropy for object classification to help improve detection accuracy.

Note: This example requires the Computer Vision Toolbox™ Model for YOLO v3 Object Detection. You can install the Computer Vision Toolbox Model for YOLO v3 Object Detection from Add-On Explorer. For more information about installing add-ons, see Get and Manage Add-Ons.

Download Pretrained Network

Download a pretrained network using the helper function downloadPretrainedYOLOv3Detector to avoid having to wait for training to complete. If you want to train the network, set the doTraining variable to true.

doTraining = true;

if ~doTraining
    preTrainedDetector = downloadPretrainedYOLOv3Detector();    
end

Load Data

This example uses a small labeled data set that contains 295 images. Many of these images come from the Caltech Cars 1999 and 2001 data sets, available at the Caltech Computational Vision website, created by Pietro Perona and used with permission. Each image contains one or two labeled instances of a vehicle. A small data set is useful for exploring the YOLO v3 training procedure, but in practice, more labeled images are needed to train a robust network.

Unzip the vehicle images and load the vehicle ground truth data.

unzip vehicleDatasetImages.zip
data = load('vehicleDatasetGroundTruth.mat');
vehicleDataset = data.vehicleDataset;

% Add the full path to the local vehicle data folder.
vehicleDataset.imageFilename = fullfile(pwd, vehicleDataset.imageFilename);

Note: In case of multiple classes, the data can also organized as three columns where the first column contains the image file names with paths, the second column contains the bounding boxes and the third column must be a cell vector that contains the label names corresponding to each bounding box. For more information on how to arrange the bounding boxes and labels, see boxLabelDatastore (Computer Vision Toolbox).

All the bounding boxes must be in the form [x y width height]. This vector specifies the upper left corner and the size of the bounding box in pixels.

Split the data set into a training set for training the network, and a test set for evaluating the network. Use 60% of the data for training set and the rest for the test set.

rng(0);
shuffledIndices = randperm(height(vehicleDataset));
idx = floor(0.6 * length(shuffledIndices));
trainingDataTbl = vehicleDataset(shuffledIndices(1:idx), :);
testDataTbl = vehicleDataset(shuffledIndices(idx+1:end), :);

Create an image datastore for loading the images.

imdsTrain = imageDatastore(trainingDataTbl.imageFilename);
imdsTest = imageDatastore(testDataTbl.imageFilename);

Create a datastore for the ground truth bounding boxes.

bldsTrain = boxLabelDatastore(trainingDataTbl(:, 2:end));
bldsTest = boxLabelDatastore(testDataTbl(:, 2:end));

Combine the image and box label datastores.

trainingData = combine(imdsTrain, bldsTrain);
testData = combine(imdsTest, bldsTest);

Use validateInputData to detect invalid images, bounding boxes or labels i.e.,

• Samples with invalid image format or containing NaNs

• Bounding boxes containing zeros/NaNs/Infs/empty

• Missing/non-categorical labels.

The values of the bounding boxes should be finite, positive, non-fractional, non-NaN and should be within the image boundary with a positive height and width. Any invalid samples must either be discarded or fixed for proper training.

validateInputData(trainingData);
validateInputData(testData);

Data Augmentation

Data augmentation is used to improve network accuracy by randomly transforming the original data during training. By using data augmentation, you can add more variety to the training data without actually having to increase the number of labeled training samples.

Use transform function to apply custom data augmentations to the training data. The augmentData helper function, listed at the end of the example, applies the following augmentations to the input data.

• Color jitter augmentation in HSV space

• Random horizontal flip

• Random scaling by 10 percent

augmentedTrainingData = transform(trainingData, @augmentData);

Read the same image four times and display the augmented training data.

% Visualize the augmented images.
augmentedData = cell(4,1);
for k = 1:4
    data = read(augmentedTrainingData);
    augmentedData{k} = insertShape(data{1,1}, 'Rectangle', data{1,2});
    reset(augmentedTrainingData);
end
figure
montage(augmentedData, 'BorderSize', 10)

Define YOLO v3 Object Detector

The YOLO v3 detector in this example is based on SqueezeNet, and uses the feature extraction network in SqueezeNet with the addition of two detection heads at the end. The second detection head is twice the size of the first detection head, so it is better able to detect small objects. Note that you can specify any number of detection heads of different sizes based on the size of the objects that you want to detect. The YOLO v3 detector uses anchor boxes estimated using training data to have better initial priors corresponding to the type of data set and to help the detector learn to predict the boxes accurately. For information about anchor boxes, see Anchor Boxes for Object Detection (Computer Vision Toolbox).

The YOLO v3 network present in the YOLO v3 detector is illustrated in the following diagram.

You can use Deep Network Designer to create the network shown in the diagram.

Specify the network input size. When choosing the network input size, consider the minimum size required to run the network itself, the size of the training images, and the computational cost incurred by processing data at the selected size. When feasible, choose a network input size that is close to the size of the training image and larger than the input size required for the network. To reduce the computational cost of running the example, specify a network input size of [227 227 3].

networkInputSize = [227 227 3];

First, use transform to preprocess the training data for computing the anchor boxes, as the training images used in this example are bigger than 227-by-227 and vary in size. Specify the number of anchors as 6 to achieve a good tradeoff between number of anchors and mean IoU. Use the estimateAnchorBoxes function to estimate the anchor boxes. For details on estimating anchor boxes, see Estimate Anchor Boxes From Training Data (Computer Vision Toolbox). In case of using a pretrained YOLOv3 object detector, the anchor boxes calculated on that particular training dataset need to be specified. Note that the estimation process is not deterministic. To prevent the estimated anchor boxes from changing while tuning other hyperparameters set the random seed prior to estimation using rng.

rng(0)
trainingDataForEstimation = transform(trainingData, @(data)preprocessData(data, networkInputSize));
numAnchors = 6;
[anchors, meanIoU] = estimateAnchorBoxes(trainingDataForEstimation, numAnchors)

anchors = 6×2

    42    34
   161   130
    97    93
   143   124
    33    24
    69    66

meanIoU = 0.8423

Specify anchorBoxes to use in both the detection heads. anchorBoxes is a cell array of [Mx1], where M denotes the number of detection heads. Each detection head consists of a [Nx2] matrix of anchors, where N is the number of anchors to use. Select anchorBoxes for each detection head based on the feature map size. Use larger anchors at lower scale and smaller anchors at higher scale. To do so, sort the anchors with the larger anchor boxes first and assign the first three to the first detection head and the next three to the second detection head.

area = anchors(:, 1).*anchors(:, 2);
[~, idx] = sort(area, 'descend');
anchors = anchors(idx, :);
anchorBoxes = {anchors(1:3,:)
    anchors(4:6,:)
    };

Load the SqueezeNet network pretrained on Imagenet data set and then specify the class names. You can also choose to load a different pretrained network trained on COCO data set such as tiny-yolov3-coco or darknet53-coco or Imagenet data set such as MobileNet-v2 or ResNet-18. YOLO v3 performs better and trains faster when you use a pretrained network.

baseNetwork = squeezenet;
classNames = trainingDataTbl.Properties.VariableNames(2:end);

Next, create the yolov3ObjectDetector object by adding the detection network soruce. Choosing the optimal detection network source requires trial and error, and you can use analyzeNetwork to find the names of potential detection network source within a network. For this example, use the fire9-concat and fire5-concat layers as DetectionNetworkSource.

yolov3Detector = yolov3ObjectDetector(baseNetwork, classNames, anchorBoxes, 'DetectionNetworkSource', {'fire9-concat', 'fire5-concat'});

Alternatively, instead of the network created above using SqueezeNet, other pretrained YOLOv3 architectures trained using larger datasets like MS-COCO can be used to transfer learn the detector on custom object detection task. Transfer learning can be realized by changing the classNames and anchorBoxes. Transfer learning workflow is recommended if the class of the custom object detection is present either as one of the class or subclass of classes trained in the pretrained network.

Preprocess Training Data

Preprocess the augmented training data to prepare for training. The preprocess (Computer Vision Toolbox) method in yolov3ObjectDetector (Computer Vision Toolbox), applies the following preprocessing operations to the input data.

• Resize the images to the network input size by maintaining the aspect ratio.

• Scale the image pixels in the range [0 1].

preprocessedTrainingData = transform(augmentedTrainingData, @(data)preprocess(yolov3Detector, data));

Read the preprocessed training data.

data = read(preprocessedTrainingData);

Display the image with the bounding boxes.

I = data{1,1};
bbox = data{1,2};
annotatedImage = insertShape(I, 'Rectangle', bbox);
annotatedImage = imresize(annotatedImage,2);
figure
imshow(annotatedImage)

Reset the datastore.

reset(preprocessedTrainingData);

Specify Training Options

Specify these training options.

• Set the number of epochs to be 80.

• Set the mini batch size as 8. Stable training can be possible with higher learning rates when higher mini batch size is used. Although, this should be set depending on the available memory.

• Set the learning rate to 0.001.

• Set the warmup period as 1000 iterations. This parameter denotes the number of iterations to increase the learning rate exponentially based on the formula $$\text{learningRate}\times {\left(\frac{\text{iteration}}{\text{warmupPeriod}}\right)}^4$$. It helps in stabilizing the gradients at higher learning rates.

• Set the L2 regularization factor to 0.0005.

• Specify the penalty threshold as 0.5. Detections that overlap less than 0.5 with the ground truth are penalized.

• Initialize the velocity of gradient as []. This is used by SGDM to store the velocity of gradients.

numEpochs = 80;
miniBatchSize = 8;
learningRate = 0.001;
warmupPeriod = 1000;
l2Regularization = 0.0005;
penaltyThreshold = 0.5;
velocity = [];

Train Model

Train on a GPU, if one is available. Using a GPU requires Parallel Computing Toolbox™ and a CUDA® enabled NVIDIA® GPU. For information about the supported compute capabilities, see GPU Support by Release (Parallel Computing Toolbox).

Use the minibatchqueue function to split the preprocessed training data into batches with the supporting function createBatchData which returns the batched images and bounding boxes combined with the respective class IDs. For faster extraction of the batch data for training, dispatchInBackground should be set to "true" which ensures the usage of parallel pool.

minibatchqueue automatically detects the availability of a GPU. If you do not have a GPU, or do not want to use one for training, set the OutputEnvironment parameter to "cpu".

if canUseParallelPool
   dispatchInBackground = true;
else
   dispatchInBackground = false;
end

mbqTrain = minibatchqueue(preprocessedTrainingData, 2,...
        "MiniBatchSize", miniBatchSize,...
        "MiniBatchFcn", @(images, boxes, labels) createBatchData(images, boxes, labels, classNames), ...
        "MiniBatchFormat", ["SSCB", ""],...
        "DispatchInBackground", dispatchInBackground,...
        "OutputCast", ["", "double"]);

Create the training progress plotter using supporting function configureTrainingProgressPlotter to see the plot while training the detector object with a custom training loop.

Finally, specify the custom training loop. For each iteration:

• Read data from the minibatchqueue. If it doesn't have any more data, reset the minibatchqueue and shuffle.

• Evaluate the model gradients using dlfeval and the modelGradients function. The function modelGradients, listed as a supporting function, returns the gradients of the loss with respect to the learnable parameters in net, the corresponding mini-batch loss, and the state of the current batch.

• Apply a weight decay factor to the gradients to regularization for more robust training.

• Determine the learning rate based on the iterations using the piecewiseLearningRateWithWarmup supporting function.

• Update the detector parameters using the sgdmupdate function.

• Update the state parameters of detector with the moving average.

• Display the learning rate, total loss, and the individual losses (box loss, object loss and class loss) for every iteration. These can be used to interpret how the respective losses are changing in each iteration. For example, a sudden spike in the box loss after few iterations implies that there are Inf or NaNs in the predictions.

• Update the training progress plot.

The training can also be terminated if the loss has saturated for few epochs.

if doTraining
    
    % Create subplots for the learning rate and mini-batch loss.
    fig = figure;
    [lossPlotter, learningRatePlotter] = configureTrainingProgressPlotter(fig);

    iteration = 0;
    % Custom training loop.
    for epoch = 1:numEpochs
          
        reset(mbqTrain);
        shuffle(mbqTrain);
        
        while(hasdata(mbqTrain))
            iteration = iteration + 1;
           
            [XTrain, YTrain] = next(mbqTrain);
            
            % Evaluate the model gradients and loss using dlfeval and the
            % modelGradients function.
            [gradients, state, lossInfo] = dlfeval(@modelGradients, yolov3Detector, XTrain, YTrain, penaltyThreshold);
    
            % Apply L2 regularization.
            gradients = dlupdate(@(g,w) g + l2Regularization*w, gradients, yolov3Detector.Learnables);
    
            % Determine the current learning rate value.
            currentLR = piecewiseLearningRateWithWarmup(iteration, epoch, learningRate, warmupPeriod, numEpochs);
    
            % Update the detector learnable parameters using the SGDM optimizer.
            [yolov3Detector.Learnables, velocity] = sgdmupdate(yolov3Detector.Learnables, gradients, velocity, currentLR);
    
            % Update the state parameters of dlnetwork.
            yolov3Detector.State = state;
              
            % Display progress.
            displayLossInfo(epoch, iteration, currentLR, lossInfo);  
                
            % Update training plot with new points.
            updatePlots(lossPlotter, learningRatePlotter, iteration, currentLR, lossInfo.totalLoss);
        end        
    end
else
    yolov3Detector = preTrainedDetector;
end

Epoch : 1 | Iteration : 1 | Learning Rate : 1e-15 | Total Loss : 2034.4574 | Box Loss : 1.2703 | Object Loss : 2032.5195 | Class Loss : 0.66761
Epoch : 1 | Iteration : 2 | Learning Rate : 1.6e-14 | Total Loss : 2040.5183 | Box Loss : 5.8543 | Object Loss : 2033.9915 | Class Loss : 0.67264
Epoch : 1 | Iteration : 3 | Learning Rate : 8.1e-14 | Total Loss : 2039.3861 | Box Loss : 5.0515 | Object Loss : 2033.4391 | Class Loss : 0.89554
Epoch : 1 | Iteration : 4 | Learning Rate : 2.56e-13 | Total Loss : 2045.5454 | Box Loss : 2.5824 | Object Loss : 2042.353 | Class Loss : 0.61002
Epoch : 1 | Iteration : 5 | Learning Rate : 6.25e-13 | Total Loss : 2034.3147 | Box Loss : 4.6497 | Object Loss : 2028.9595 | Class Loss : 0.70542
Epoch : 1 | Iteration : 6 | Learning Rate : 1.296e-12 | Total Loss : 2038.448 | Box Loss : 6.7472 | Object Loss : 2031.0361 | Class Loss : 0.66469
Epoch : 1 | Iteration : 7 | Learning Rate : 2.401e-12 | Total Loss : 2043.5757 | Box Loss : 2.474 | Object Loss : 2040.5457 | Class Loss : 0.55609
Epoch : 1 | Iteration : 8 | Learning Rate : 4.096e-12 | Total Loss : 2044.3937 | Box Loss : 6.5438 | Object Loss : 2037.5107 | Class Loss : 0.33914
Epoch : 1 | Iteration : 9 | Learning Rate : 6.561e-12 | Total Loss : 2029.511 | Box Loss : 2.3748 | Object Loss : 2026.4377 | Class Loss : 0.69853
Epoch : 1 | Iteration : 10 | Learning Rate : 1e-11 | Total Loss : 2026.5184 | Box Loss : 2.2127 | Object Loss : 2023.8136 | Class Loss : 0.49224
Epoch : 1 | Iteration : 11 | Learning Rate : 1.4641e-11 | Total Loss : 2052.4109 | Box Loss : 4.4924 | Object Loss : 2047.2883 | Class Loss : 0.63001
Epoch : 1 | Iteration : 12 | Learning Rate : 2.0736e-11 | Total Loss : 2039.0267 | Box Loss : 4.2858 | Object Loss : 2034.1895 | Class Loss : 0.55147
Epoch : 1 | Iteration : 13 | Learning Rate : 2.8561e-11 | Total Loss : 2053.4932 | Box Loss : 2.1127 | Object Loss : 2050.6885 | Class Loss : 0.69185
Epoch : 1 | Iteration : 14 | Learning Rate : 3.8416e-11 | Total Loss : 2040.917 | Box Loss : 2.8612 | Object Loss : 2037.4712 | Class Loss : 0.58456
Epoch : 1 | Iteration : 15 | Learning Rate : 5.0625e-11 | Total Loss : 2043.094 | Box Loss : 2.8008 | Object Loss : 2039.9056 | Class Loss : 0.3876
Epoch : 1 | Iteration : 16 | Learning Rate : 6.5536e-11 | Total Loss : 2031.8059 | Box Loss : 3.2756 | Object Loss : 2028.0403 | Class Loss : 0.49002
Epoch : 1 | Iteration : 17 | Learning Rate : 8.3521e-11 | Total Loss : 2044.72 | Box Loss : 1.6522 | Object Loss : 2042.4524 | Class Loss : 0.61532
Epoch : 1 | Iteration : 18 | Learning Rate : 1.0498e-10 | Total Loss : 2050.0471 | Box Loss : 4.1119 | Object Loss : 2045.4639 | Class Loss : 0.47138
Epoch : 1 | Iteration : 19 | Learning Rate : 1.3032e-10 | Total Loss : 2033.7067 | Box Loss : 2.8427 | Object Loss : 2030.1394 | Class Loss : 0.72465
Epoch : 1 | Iteration : 20 | Learning Rate : 1.6e-10 | Total Loss : 2036.5347 | Box Loss : 3.1793 | Object Loss : 2032.9583 | Class Loss : 0.39718
Epoch : 1 | Iteration : 21 | Learning Rate : 1.9448e-10 | Total Loss : 2029.7346 | Box Loss : 3.6625 | Object Loss : 2025.3025 | Class Loss : 0.76967
Epoch : 1 | Iteration : 22 | Learning Rate : 2.3426e-10 | Total Loss : 2026.9553 | Box Loss : 2.9675 | Object Loss : 2023.5861 | Class Loss : 0.4017
Epoch : 1 | Iteration : 23 | Learning Rate : 2.7984e-10 | Total Loss : 2019.0176 | Box Loss : 0.84919 | Object Loss : 2017.3557 | Class Loss : 0.81262
Epoch : 2 | Iteration : 24 | Learning Rate : 3.3178e-10 | Total Loss : 2033.5848 | Box Loss : 3.0358 | Object Loss : 2029.8921 | Class Loss : 0.65696
Epoch : 2 | Iteration : 25 | Learning Rate : 3.9063e-10 | Total Loss : 2045.8374 | Box Loss : 3.4959 | Object Loss : 2041.6012 | Class Loss : 0.74038
Epoch : 2 | Iteration : 26 | Learning Rate : 4.5698e-10 | Total Loss : 2036.3354 | Box Loss : 2.1746 | Object Loss : 2033.5701 | Class Loss : 0.59084
Epoch : 2 | Iteration : 27 | Learning Rate : 5.3144e-10 | Total Loss : 2036.7156 | Box Loss : 3.0053 | Object Loss : 2032.9835 | Class Loss : 0.72679
Epoch : 2 | Iteration : 28 | Learning Rate : 6.1466e-10 | Total Loss : 2030.1866 | Box Loss : 3.4409 | Object Loss : 2026.0948 | Class Loss : 0.65091
Epoch : 2 | Iteration : 29 | Learning Rate : 7.0728e-10 | Total Loss : 2026.7745 | Box Loss : 1.0128 | Object Loss : 2025.1301 | Class Loss : 0.63154
Epoch : 2 | Iteration : 30 | Learning Rate : 8.1e-10 | Total Loss : 2039.3251 | Box Loss : 3.1312 | Object Loss : 2035.4562 | Class Loss : 0.73767
Epoch : 2 | Iteration : 31 | Learning Rate : 9.2352e-10 | Total Loss : 2034.1394 | Box Loss : 4.8098 | Object Loss : 2028.5234 | Class Loss : 0.8062
Epoch : 2 | Iteration : 32 | Learning Rate : 1.0486e-09 | Total Loss : 2035.0363 | Box Loss : 4.7082 | Object Loss : 2029.7371 | Class Loss : 0.59096
Epoch : 2 | Iteration : 33 | Learning Rate : 1.1859e-09 | Total Loss : 2053.9387 | Box Loss : 3.5839 | Object Loss : 2049.7886 | Class Loss : 0.56609
Epoch : 2 | Iteration : 34 | Learning Rate : 1.3363e-09 | Total Loss : 2041.5179 | Box Loss : 2.808 | Object Loss : 2038.3765 | Class Loss : 0.33344
Epoch : 2 | Iteration : 35 | Learning Rate : 1.5006e-09 | Total Loss : 2035.2411 | Box Loss : 2.7223 | Object Loss : 2032.0865 | Class Loss : 0.43231
Epoch : 2 | Iteration : 36 | Learning Rate : 1.6796e-09 | Total Loss : 2050.2747 | Box Loss : 5.3999 | Object Loss : 2044.2727 | Class Loss : 0.60193
Epoch : 2 | Iteration : 37 | Learning Rate : 1.8742e-09 | Total Loss : 2043.5969 | Box Loss : 5.5765 | Object Loss : 2037.3926 | Class Loss : 0.6279
Epoch : 2 | Iteration : 38 | Learning Rate : 2.0851e-09 | Total Loss : 2038.2933 | Box Loss : 3.4637 | Object Loss : 2034.1479 | Class Loss : 0.6816
Epoch : 2 | Iteration : 39 | Learning Rate : 2.3134e-09 | Total Loss : 2038.1877 | Box Loss : 2.5275 | Object Loss : 2035.101 | Class Loss : 0.55933
Epoch : 2 | Iteration : 40 | Learning Rate : 2.56e-09 | Total Loss : 2036.8016 | Box Loss : 2.0805 | Object Loss : 2034.0248 | Class Loss : 0.69634
Epoch : 2 | Iteration : 41 | Learning Rate : 2.8258e-09 | Total Loss : 2032.4956 | Box Loss : 2.0393 | Object Loss : 2030.0411 | Class Loss : 0.41518
Epoch : 2 | Iteration : 42 | Learning Rate : 3.1117e-09 | Total Loss : 2036.3785 | Box Loss : 3.8184 | Object Loss : 2032.1405 | Class Loss : 0.41965
Epoch : 2 | Iteration : 43 | Learning Rate : 3.4188e-09 | Total Loss : 2037.4792 | Box Loss : 2.2771 | Object Loss : 2034.524 | Class Loss : 0.67814
Epoch : 2 | Iteration : 44 | Learning Rate : 3.7481e-09 | Total Loss : 2036.874 | Box Loss : 2.9961 | Object Loss : 2033.1942 | Class Loss : 0.68376
Epoch : 2 | Iteration : 45 | Learning Rate : 4.1006e-09 | Total Loss : 2042.7714 | Box Loss : 5.42 | Object Loss : 2036.751 | Class Loss : 0.60037
Epoch : 2 | Iteration : 46 | Learning Rate : 4.4775e-09 | Total Loss : 2040.7478 | Box Loss : 15.8955 | Object Loss : 2024.0149 | Class Loss : 0.83739
Epoch : 3 | Iteration : 47 | Learning Rate : 4.8797e-09 | Total Loss : 2029.9019 | Box Loss : 5.3115 | Object Loss : 2023.8076 | Class Loss : 0.78269
Epoch : 3 | Iteration : 48 | Learning Rate : 5.3084e-09 | Total Loss : 2037.0276 | Box Loss : 3.8952 | Object Loss : 2032.524 | Class Loss : 0.60828
Epoch : 3 | Iteration : 49 | Learning Rate : 5.7648e-09 | Total Loss : 2024.6521 | Box Loss : 2.2851 | Object Loss : 2021.6416 | Class Loss : 0.72537
Epoch : 3 | Iteration : 50 | Learning Rate : 6.25e-09 | Total Loss : 2030.0686 | Box Loss : 3.7038 | Object Loss : 2025.936 | Class Loss : 0.42867
Epoch : 3 | Iteration : 51 | Learning Rate : 6.7652e-09 | Total Loss : 2033.3076 | Box Loss : 1.6001 | Object Loss : 2031.0587 | Class Loss : 0.64882
Epoch : 3 | Iteration : 52 | Learning Rate : 7.3116e-09 | Total Loss : 2031.2271 | Box Loss : 2.8679 | Object Loss : 2027.5931 | Class Loss : 0.76597
Epoch : 3 | Iteration : 53 | Learning Rate : 7.8905e-09 | Total Loss : 2026.3621 | Box Loss : 1.5324 | Object Loss : 2024.1782 | Class Loss : 0.65154
Epoch : 3 | Iteration : 54 | Learning Rate : 8.5031e-09 | Total Loss : 2031.1776 | Box Loss : 3.7035 | Object Loss : 2026.9082 | Class Loss : 0.56593
Epoch : 3 | Iteration : 55 | Learning Rate : 9.1506e-09 | Total Loss : 2034.3107 | Box Loss : 2.6567 | Object Loss : 2030.9513 | Class Loss : 0.70268
Epoch : 3 | Iteration : 56 | Learning Rate : 9.8345e-09 | Total Loss : 2021.8693 | Box Loss : 3.34 | Object Loss : 2017.9197 | Class Loss : 0.60955
Epoch : 3 | Iteration : 57 | Learning Rate : 1.0556e-08 | Total Loss : 2024.7795 | Box Loss : 5.4342 | Object Loss : 2018.5682 | Class Loss : 0.77708
Epoch : 3 | Iteration : 58 | Learning Rate : 1.1316e-08 | Total Loss : 2021.3302 | Box Loss : 2.4311 | Object Loss : 2018.3871 | Class Loss : 0.5121
Epoch : 3 | Iteration : 59 | Learning Rate : 1.2117e-08 | Total Loss : 2028.3229 | Box Loss : 3.0344 | Object Loss : 2024.6876 | Class Loss : 0.60097
Epoch : 3 | Iteration : 60 | Learning Rate : 1.296e-08 | Total Loss : 2029.2771 | Box Loss : 3.7049 | Object Loss : 2025.1646 | Class Loss : 0.40764
Epoch : 3 | Iteration : 61 | Learning Rate : 1.3846e-08 | Total Loss : 2023.4316 | Box Loss : 2.836 | Object Loss : 2020.1001 | Class Loss : 0.49552
Epoch : 3 | Iteration : 62 | Learning Rate : 1.4776e-08 | Total Loss : 2018.3862 | Box Loss : 3.4428 | Object Loss : 2014.299 | Class Loss : 0.64438
Epoch : 3 | Iteration : 63 | Learning Rate : 1.5753e-08 | Total Loss : 1997.1346 | Box Loss : 5.3684 | Object Loss : 1991.0413 | Class Loss : 0.72492
Epoch : 3 | Iteration : 64 | Learning Rate : 1.6777e-08 | Total Loss : 2008.8459 | Box Loss : 3.0047 | Object Loss : 2005.2698 | Class Loss : 0.57158
Epoch : 3 | Iteration : 65 | Learning Rate : 1.7851e-08 | Total Loss : 2003.0519 | Box Loss : 3.0569 | Object Loss : 1999.3745 | Class Loss : 0.62048
Epoch : 3 | Iteration : 66 | Learning Rate : 1.8975e-08 | Total Loss : 1996.9225 | Box Loss : 5.2107 | Object Loss : 1991.1674 | Class Loss : 0.54448
Epoch : 3 | Iteration : 67 | Learning Rate : 2.0151e-08 | Total Loss : 1993.1699 | Box Loss : 2.067 | Object Loss : 1990.5317 | Class Loss : 0.5712
Epoch : 3 | Iteration : 68 | Learning Rate : 2.1381e-08 | Total Loss : 1996.2117 | Box Loss : 1.808 | Object Loss : 1993.8293 | Class Loss : 0.57438
Epoch : 3 | Iteration : 69 | Learning Rate : 2.2667e-08 | Total Loss : 1960.97 | Box Loss : 3.5167 | Object Loss : 1957.0793 | Class Loss : 0.37408
Epoch : 4 | Iteration : 70 | Learning Rate : 2.401e-08 | Total Loss : 1987.9667 | Box Loss : 3.758 | Object Loss : 1983.6907 | Class Loss : 0.518
Epoch : 4 | Iteration : 71 | Learning Rate : 2.5412e-08 | Total Loss : 1985.2295 | Box Loss : 1.5088 | Object Loss : 1983.1033 | Class Loss : 0.6174
Epoch : 4 | Iteration : 72 | Learning Rate : 2.6874e-08 | Total Loss : 1986.2605 | Box Loss : 5.0329 | Object Loss : 1980.4592 | Class Loss : 0.76839
Epoch : 4 | Iteration : 73 | Learning Rate : 2.8398e-08 | Total Loss : 1983.7778 | Box Loss : 2.5958 | Object Loss : 1980.4906 | Class Loss : 0.69158
Epoch : 4 | Iteration : 74 | Learning Rate : 2.9987e-08 | Total Loss : 1983.8669 | Box Loss : 3.2597 | Object Loss : 1980.2219 | Class Loss : 0.38521
Epoch : 4 | Iteration : 75 | Learning Rate : 3.1641e-08 | Total Loss : 1965.298 | Box Loss : 2.8892 | Object Loss : 1962.0405 | Class Loss : 0.3683
Epoch : 4 | Iteration : 76 | Learning Rate : 3.3362e-08 | Total Loss : 1975.8003 | Box Loss : 4.5828 | Object Loss : 1970.3978 | Class Loss : 0.81973
Epoch : 4 | Iteration : 77 | Learning Rate : 3.5153e-08 | Total Loss : 1956.8281 | Box Loss : 3.4007 | Object Loss : 1952.7102 | Class Loss : 0.71722
Epoch : 4 | Iteration : 78 | Learning Rate : 3.7015e-08 | Total Loss : 1947.5746 | Box Loss : 3.6963 | Object Loss : 1943.3844 | Class Loss : 0.49393
Epoch : 4 | Iteration : 79 | Learning Rate : 3.895e-08 | Total Loss : 1945.3359 | Box Loss : 2.7079 | Object Loss : 1941.9802 | Class Loss : 0.64784
Epoch : 4 | Iteration : 80 | Learning Rate : 4.096e-08 | Total Loss : 1936.6976 | Box Loss : 3.5695 | Object Loss : 1932.4708 | Class Loss : 0.65735
Epoch : 4 | Iteration : 81 | Learning Rate : 4.3047e-08 | Total Loss : 1940.713 | Box Loss : 3.8212 | Object Loss : 1936.5292 | Class Loss : 0.36271
Epoch : 4 | Iteration : 82 | Learning Rate : 4.5212e-08 | Total Loss : 1920.8802 | Box Loss : 3.2441 | Object Loss : 1917.2192 | Class Loss : 0.41683
Epoch : 4 | Iteration : 83 | Learning Rate : 4.7458e-08 | Total Loss : 1911.2277 | Box Loss : 3.8905 | Object Loss : 1906.6838 | Class Loss : 0.65335
Epoch : 4 | Iteration : 84 | Learning Rate : 4.9787e-08 | Total Loss : 1911.212 | Box Loss : 3.6702 | Object Loss : 1906.9504 | Class Loss : 0.59146
Epoch : 4 | Iteration : 85 | Learning Rate : 5.2201e-08 | Total Loss : 1890.0442 | Box Loss : 2.015 | Object Loss : 1887.3788 | Class Loss : 0.65035
Epoch : 4 | Iteration : 86 | Learning Rate : 5.4701e-08 | Total Loss : 1894.0687 | Box Loss : 2.3115 | Object Loss : 1891.0864 | Class Loss : 0.67088
Epoch : 4 | Iteration : 87 | Learning Rate : 5.729e-08 | Total Loss : 1882.7527 | Box Loss : 3.0078 | Object Loss : 1878.8282 | Class Loss : 0.91661
Epoch : 4 | Iteration : 88 | Learning Rate : 5.997e-08 | Total Loss : 1878.7745 | Box Loss : 4.5958 | Object Loss : 1873.644 | Class Loss : 0.53463
Epoch : 4 | Iteration : 89 | Learning Rate : 6.2742e-08 | Total Loss : 1874.4493 | Box Loss : 3.7955 | Object Loss : 1870.1482 | Class Loss : 0.50562
Epoch : 4 | Iteration : 90 | Learning Rate : 6.561e-08 | Total Loss : 1849.9515 | Box Loss : 3.1729 | Object Loss : 1846.0564 | Class Loss : 0.72234
Epoch : 4 | Iteration : 91 | Learning Rate : 6.8575e-08 | Total Loss : 1832.1216 | Box Loss : 1.972 | Object Loss : 1829.5789 | Class Loss : 0.57063
Epoch : 4 | Iteration : 92 | Learning Rate : 7.1639e-08 | Total Loss : 1825.4591 | Box Loss : 16.0346 | Object Loss : 1808.7312 | Class Loss : 0.69319
Epoch : 5 | Iteration : 93 | Learning Rate : 7.4805e-08 | Total Loss : 1822.5862 | Box Loss : 3.3594 | Object Loss : 1818.7134 | Class Loss : 0.51327
Epoch : 5 | Iteration : 94 | Learning Rate : 7.8075e-08 | Total Loss : 1806.3173 | Box Loss : 4.9844 | Object Loss : 1800.8016 | Class Loss : 0.53124
Epoch : 5 | Iteration : 95 | Learning Rate : 8.1451e-08 | Total Loss : 1791.694 | Box Loss : 2.6037 | Object Loss : 1788.7158 | Class Loss : 0.37446
Epoch : 5 | Iteration : 96 | Learning Rate : 8.4935e-08 | Total Loss : 1788.7694 | Box Loss : 3.4347 | Object Loss : 1784.615 | Class Loss : 0.71978
Epoch : 5 | Iteration : 97 | Learning Rate : 8.8529e-08 | Total Loss : 1765.125 | Box Loss : 2.1152 | Object Loss : 1762.3104 | Class Loss : 0.69946
Epoch : 5 | Iteration : 98 | Learning Rate : 9.2237e-08 | Total Loss : 1750.5256 | Box Loss : 2.0819 | Object Loss : 1747.7401 | Class Loss : 0.70358
Epoch : 5 | Iteration : 99 | Learning Rate : 9.606e-08 | Total Loss : 1743.0345 | Box Loss : 1.6052 | Object Loss : 1741.0526 | Class Loss : 0.37671
Epoch : 5 | Iteration : 100 | Learning Rate : 1e-07 | Total Loss : 1731.6486 | Box Loss : 2.262 | Object Loss : 1728.8052 | Class Loss : 0.5813
Epoch : 5 | Iteration : 101 | Learning Rate : 1.0406e-07 | Total Loss : 1721.0155 | Box Loss : 2.7817 | Object Loss : 1717.8625 | Class Loss : 0.37132
Epoch : 5 | Iteration : 102 | Learning Rate : 1.0824e-07 | Total Loss : 1703.8527 | Box Loss : 1.5948 | Object Loss : 1701.7136 | Class Loss : 0.54421
Epoch : 5 | Iteration : 103 | Learning Rate : 1.1255e-07 | Total Loss : 1681.6267 | Box Loss : 3.4908 | Object Loss : 1677.3575 | Class Loss : 0.77836
Epoch : 5 | Iteration : 104 | Learning Rate : 1.1699e-07 | Total Loss : 1663.0557 | Box Loss : 6.277 | Object Loss : 1656.0092 | Class Loss : 0.76947
Epoch : 5 | Iteration : 105 | Learning Rate : 1.2155e-07 | Total Loss : 1651.8859 | Box Loss : 3.8958 | Object Loss : 1647.5018 | Class Loss : 0.48832
Epoch : 5 | Iteration : 106 | Learning Rate : 1.2625e-07 | Total Loss : 1637.6196 | Box Loss : 3.4216 | Object Loss : 1633.6617 | Class Loss : 0.53624
Epoch : 5 | Iteration : 107 | Learning Rate : 1.3108e-07 | Total Loss : 1611.4268 | Box Loss : 2.0408 | Object Loss : 1608.6917 | Class Loss : 0.69436
Epoch : 5 | Iteration : 108 | Learning Rate : 1.3605e-07 | Total Loss : 1588.9783 | Box Loss : 4.4034 | Object Loss : 1584.0145 | Class Loss : 0.56031
Epoch : 5 | Iteration : 109 | Learning Rate : 1.4116e-07 | Total Loss : 1577.9961 | Box Loss : 3.5424 | Object Loss : 1573.731 | Class Loss : 0.7226
Epoch : 5 | Iteration : 110 | Learning Rate : 1.4641e-07 | Total Loss : 1554.6068 | Box Loss : 2.9358 | Object Loss : 1551.1073 | Class Loss : 0.56367
Epoch : 5 | Iteration : 111 | Learning Rate : 1.5181e-07 | Total Loss : 1545.191 | Box Loss : 2.9433 | Object Loss : 1541.4181 | Class Loss : 0.82967
Epoch : 5 | Iteration : 112 | Learning Rate : 1.5735e-07 | Total Loss : 1516.0305 | Box Loss : 2.7912 | Object Loss : 1512.7938 | Class Loss : 0.44542
Epoch : 5 | Iteration : 113 | Learning Rate : 1.6305e-07 | Total Loss : 1504.9351 | Box Loss : 6.1784 | Object Loss : 1498.2 | Class Loss : 0.55669
Epoch : 5 | Iteration : 114 | Learning Rate : 1.689e-07 | Total Loss : 1481.5167 | Box Loss : 3.3483 | Object Loss : 1477.4575 | Class Loss : 0.71095
Epoch : 5 | Iteration : 115 | Learning Rate : 1.749e-07 | Total Loss : 1446.4066 | Box Loss : 4.169 | Object Loss : 1441.9678 | Class Loss : 0.26987
Epoch : 6 | Iteration : 116 | Learning Rate : 1.8106e-07 | Total Loss : 1447.542 | Box Loss : 4.7388 | Object Loss : 1442.2832 | Class Loss : 0.52001
Epoch : 6 | Iteration : 117 | Learning Rate : 1.8739e-07 | Total Loss : 1410.4972 | Box Loss : 4.8979 | Object Loss : 1404.7754 | Class Loss : 0.8239
Epoch : 6 | Iteration : 118 | Learning Rate : 1.9388e-07 | Total Loss : 1401.3512 | Box Loss : 2.5921 | Object Loss : 1398.2458 | Class Loss : 0.51321
Epoch : 6 | Iteration : 119 | Learning Rate : 2.0053e-07 | Total Loss : 1370.1278 | Box Loss : 1.8787 | Object Loss : 1367.7649 | Class Loss : 0.48423
Epoch : 6 | Iteration : 120 | Learning Rate : 2.0736e-07 | Total Loss : 1352.739 | Box Loss : 4.8197 | Object Loss : 1347.353 | Class Loss : 0.56634
Epoch : 6 | Iteration : 121 | Learning Rate : 2.1436e-07 | Total Loss : 1333.5609 | Box Loss : 2.1018 | Object Loss : 1330.7599 | Class Loss : 0.69922
Epoch : 6 | Iteration : 122 | Learning Rate : 2.2153e-07 | Total Loss : 1299.8704 | Box Loss : 2.1882 | Object Loss : 1297.166 | Class Loss : 0.51607
Epoch : 6 | Iteration : 123 | Learning Rate : 2.2889e-07 | Total Loss : 1282.7609 | Box Loss : 3.0205 | Object Loss : 1279.2159 | Class Loss : 0.52443
Epoch : 6 | Iteration : 124 | Learning Rate : 2.3642e-07 | Total Loss : 1272.4924 | Box Loss : 2.5574 | Object Loss : 1269.4204 | Class Loss : 0.51467
Epoch : 6 | Iteration : 125 | Learning Rate : 2.4414e-07 | Total Loss : 1250.6395 | Box Loss : 4.9803 | Object Loss : 1244.9894 | Class Loss : 0.66982
Epoch : 6 | Iteration : 126 | Learning Rate : 2.5205e-07 | Total Loss : 1215.4714 | Box Loss : 4.1274 | Object Loss : 1210.7234 | Class Loss : 0.62054
Epoch : 6 | Iteration : 127 | Learning Rate : 2.6014e-07 | Total Loss : 1198.6125 | Box Loss : 3.395 | Object Loss : 1194.76 | Class Loss : 0.45757
Epoch : 6 | Iteration : 128 | Learning Rate : 2.6844e-07 | Total Loss : 1166.0612 | Box Loss : 1.8501 | Object Loss : 1163.7635 | Class Loss : 0.44757
Epoch : 6 | Iteration : 129 | Learning Rate : 2.7692e-07 | Total Loss : 1145.3256 | Box Loss : 5.843 | Object Loss : 1138.7683 | Class Loss : 0.71427
Epoch : 6 | Iteration : 130 | Learning Rate : 2.8561e-07 | Total Loss : 1121.3236 | Box Loss : 1.5061 | Object Loss : 1119.2224 | Class Loss : 0.59509
Epoch : 6 | Iteration : 131 | Learning Rate : 2.945e-07 | Total Loss : 1107.7411 | Box Loss : 1.7676 | Object Loss : 1105.3784 | Class Loss : 0.59508
Epoch : 6 | Iteration : 132 | Learning Rate : 3.036e-07 | Total Loss : 1084.3871 | Box Loss : 3.8119 | Object Loss : 1079.9816 | Class Loss : 0.59367
Epoch : 6 | Iteration : 133 | Learning Rate : 3.129e-07 | Total Loss : 1061.6428 | Box Loss : 2.3155 | Object Loss : 1058.6475 | Class Loss : 0.67995
Epoch : 6 | Iteration : 134 | Learning Rate : 3.2242e-07 | Total Loss : 1032.1377 | Box Loss : 2.9454 | Object Loss : 1028.4749 | Class Loss : 0.71751
Epoch : 6 | Iteration : 135 | Learning Rate : 3.3215e-07 | Total Loss : 1010.8308 | Box Loss : 5.0547 | Object Loss : 1005.1586 | Class Loss : 0.61748
Epoch : 6 | Iteration : 136 | Learning Rate : 3.421e-07 | Total Loss : 980.6289 | Box Loss : 2.3295 | Object Loss : 977.5277 | Class Loss : 0.77165
Epoch : 6 | Iteration : 137 | Learning Rate : 3.5228e-07 | Total Loss : 954.4264 | Box Loss : 1.2685 | Object Loss : 952.7402 | Class Loss : 0.41771
Epoch : 6 | Iteration : 138 | Learning Rate : 3.6267e-07 | Total Loss : 950.8206 | Box Loss : 5.479 | Object Loss : 944.7975 | Class Loss : 0.54406
Epoch : 7 | Iteration : 139 | Learning Rate : 3.733e-07 | Total Loss : 920.5997 | Box Loss : 3.7647 | Object Loss : 916.1279 | Class Loss : 0.70721
Epoch : 7 | Iteration : 140 | Learning Rate : 3.8416e-07 | Total Loss : 886.7099 | Box Loss : 2.1462 | Object Loss : 884.063 | Class Loss : 0.50076
Epoch : 7 | Iteration : 141 | Learning Rate : 3.9525e-07 | Total Loss : 874.2281 | Box Loss : 3.4884 | Object Loss : 870.3276 | Class Loss : 0.41216
Epoch : 7 | Iteration : 142 | Learning Rate : 4.0659e-07 | Total Loss : 856.9265 | Box Loss : 2.3927 | Object Loss : 854.1731 | Class Loss : 0.3606
Epoch : 7 | Iteration : 143 | Learning Rate : 4.1816e-07 | Total Loss : 830.4821 | Box Loss : 2.4321 | Object Loss : 827.4305 | Class Loss : 0.61948
Epoch : 7 | Iteration : 144 | Learning Rate : 4.2998e-07 | Total Loss : 803.5051 | Box Loss : 4.0953 | Object Loss : 798.4871 | Class Loss : 0.92264
Epoch : 7 | Iteration : 145 | Learning Rate : 4.4205e-07 | Total Loss : 787.3389 | Box Loss : 2.6817 | Object Loss : 784.1239 | Class Loss : 0.53327
Epoch : 7 | Iteration : 146 | Learning Rate : 4.5437e-07 | Total Loss : 766.8655 | Box Loss : 4.5345 | Object Loss : 761.6417 | Class Loss : 0.68924
Epoch : 7 | Iteration : 147 | Learning Rate : 4.6695e-07 | Total Loss : 747.8558 | Box Loss : 1.6245 | Object Loss : 745.6534 | Class Loss : 0.57786
Epoch : 7 | Iteration : 148 | Learning Rate : 4.7979e-07 | Total Loss : 724.8915 | Box Loss : 4.2896 | Object Loss : 720.0224 | Class Loss : 0.57949
Epoch : 7 | Iteration : 149 | Learning Rate : 4.9288e-07 | Total Loss : 707.7861 | Box Loss : 3.0635 | Object Loss : 704.1108 | Class Loss : 0.61181
Epoch : 7 | Iteration : 150 | Learning Rate : 5.0625e-07 | Total Loss : 682.8471 | Box Loss : 1.9388 | Object Loss : 680.3842 | Class Loss : 0.5242
Epoch : 7 | Iteration : 151 | Learning Rate : 5.1989e-07 | Total Loss : 661.091 | Box Loss : 2.2959 | Object Loss : 658.3548 | Class Loss : 0.44027
Epoch : 7 | Iteration : 152 | Learning Rate : 5.3379e-07 | Total Loss : 649.2206 | Box Loss : 2.3278 | Object Loss : 646.3533 | Class Loss : 0.53957
Epoch : 7 | Iteration : 153 | Learning Rate : 5.4798e-07 | Total Loss : 628.3187 | Box Loss : 2.3732 | Object Loss : 625.4421 | Class Loss : 0.50344
Epoch : 7 | Iteration : 154 | Learning Rate : 5.6245e-07 | Total Loss : 611.2161 | Box Loss : 1.4037 | Object Loss : 609.1978 | Class Loss : 0.61459
Epoch : 7 | Iteration : 155 | Learning Rate : 5.772e-07 | Total Loss : 592.9767 | Box Loss : 2.4407 | Object Loss : 589.9943 | Class Loss : 0.54181
Epoch : 7 | Iteration : 156 | Learning Rate : 5.9224e-07 | Total Loss : 575.8509 | Box Loss : 4.8618 | Object Loss : 570.1839 | Class Loss : 0.80518
Epoch : 7 | Iteration : 157 | Learning Rate : 6.0757e-07 | Total Loss : 558.6267 | Box Loss : 2.241 | Object Loss : 555.8766 | Class Loss : 0.50903
Epoch : 7 | Iteration : 158 | Learning Rate : 6.232e-07 | Total Loss : 545.6277 | Box Loss : 1.87 | Object Loss : 542.9493 | Class Loss : 0.8084
Epoch : 7 | Iteration : 159 | Learning Rate : 6.3913e-07 | Total Loss : 533.227 | Box Loss : 2.6881 | Object Loss : 530.0394 | Class Loss : 0.49956
Epoch : 7 | Iteration : 160 | Learning Rate : 6.5536e-07 | Total Loss : 510.7053 | Box Loss : 3.0112 | Object Loss : 507.2845 | Class Loss : 0.4097
Epoch : 7 | Iteration : 161 | Learning Rate : 6.719e-07 | Total Loss : 500.6588 | Box Loss : 1.882 | Object Loss : 498.0595 | Class Loss : 0.71724
Epoch : 8 | Iteration : 162 | Learning Rate : 6.8875e-07 | Total Loss : 482.7756 | Box Loss : 4.3151 | Object Loss : 477.9934 | Class Loss : 0.46708
Epoch : 8 | Iteration : 163 | Learning Rate : 7.0591e-07 | Total Loss : 468.5723 | Box Loss : 3.9805 | Object Loss : 463.912 | Class Loss : 0.67974
Epoch : 8 | Iteration : 164 | Learning Rate : 7.2339e-07 | Total Loss : 455.5461 | Box Loss : 2.7851 | Object Loss : 452.0892 | Class Loss : 0.67181
Epoch : 8 | Iteration : 165 | Learning Rate : 7.412e-07 | Total Loss : 440.152 | Box Loss : 1.8796 | Object Loss : 437.8831 | Class Loss : 0.3893
Epoch : 8 | Iteration : 166 | Learning Rate : 7.5933e-07 | Total Loss : 424.8647 | Box Loss : 0.95922 | Object Loss : 423.4963 | Class Loss : 0.40921
Epoch : 8 | Iteration : 167 | Learning Rate : 7.778e-07 | Total Loss : 413.4631 | Box Loss : 1.684 | Object Loss : 411.2394 | Class Loss : 0.53968
Epoch : 8 | Iteration : 168 | Learning Rate : 7.9659e-07 | Total Loss : 402.9919 | Box Loss : 3.5577 | Object Loss : 398.8732 | Class Loss : 0.56104
Epoch : 8 | Iteration : 169 | Learning Rate : 8.1573e-07 | Total Loss : 388.0714 | Box Loss : 2.3663 | Object Loss : 385.1064 | Class Loss : 0.59866
Epoch : 8 | Iteration : 170 | Learning Rate : 8.3521e-07 | Total Loss : 374.811 | Box Loss : 1.3496 | Object Loss : 373.0375 | Class Loss : 0.42392
Epoch : 8 | Iteration : 171 | Learning Rate : 8.5504e-07 | Total Loss : 365.7278 | Box Loss : 3.9867 | Object Loss : 361.0733 | Class Loss : 0.66776
Epoch : 8 | Iteration : 172 | Learning Rate : 8.7521e-07 | Total Loss : 356.4079 | Box Loss : 2.453 | Object Loss : 353.4033 | Class Loss : 0.55167
Epoch : 8 | Iteration : 173 | Learning Rate : 8.9575e-07 | Total Loss : 343.1337 | Box Loss : 2.5413 | Object Loss : 339.976 | Class Loss : 0.61644
Epoch : 8 | Iteration : 174 | Learning Rate : 9.1664e-07 | Total Loss : 334.107 | Box Loss : 2.193 | Object Loss : 331.0905 | Class Loss : 0.82353
Epoch : 8 | Iteration : 175 | Learning Rate : 9.3789e-07 | Total Loss : 328.5016 | Box Loss : 4.1864 | Object Loss : 323.8095 | Class Loss : 0.50573
Epoch : 8 | Iteration : 176 | Learning Rate : 9.5951e-07 | Total Loss : 311.0597 | Box Loss : 1.8204 | Object Loss : 308.9083 | Class Loss : 0.33105
Epoch : 8 | Iteration : 177 | Learning Rate : 9.8151e-07 | Total Loss : 307.1644 | Box Loss : 3.8359 | Object Loss : 302.7261 | Class Loss : 0.60237
Epoch : 8 | Iteration : 178 | Learning Rate : 1.0039e-06 | Total Loss : 295.8521 | Box Loss : 1.7389 | Object Loss : 293.512 | Class Loss : 0.60122
Epoch : 8 | Iteration : 179 | Learning Rate : 1.0266e-06 | Total Loss : 290.2374 | Box Loss : 2.9788 | Object Loss : 286.4553 | Class Loss : 0.80327
Epoch : 8 | Iteration : 180 | Learning Rate : 1.0498e-06 | Total Loss : 278.6443 | Box Loss : 1.6312 | Object Loss : 276.5125 | Class Loss : 0.50063
Epoch : 8 | Iteration : 181 | Learning Rate : 1.0733e-06 | Total Loss : 272.516 | Box Loss : 1.741 | Object Loss : 270.2593 | Class Loss : 0.51579
Epoch : 8 | Iteration : 182 | Learning Rate : 1.0972e-06 | Total Loss : 262.8264 | Box Loss : 2.4952 | Object Loss : 259.7936 | Class Loss : 0.53762
Epoch : 8 | Iteration : 183 | Learning Rate : 1.1215e-06 | Total Loss : 255.7794 | Box Loss : 1.3871 | Object Loss : 254.0543 | Class Loss : 0.33794
Epoch : 8 | Iteration : 184 | Learning Rate : 1.1462e-06 | Total Loss : 249.4258 | Box Loss : 4.6335 | Object Loss : 244.1304 | Class Loss : 0.6619
Epoch : 9 | Iteration : 185 | Learning Rate : 1.1714e-06 | Total Loss : 245.537 | Box Loss : 6.6451 | Object Loss : 238.3023 | Class Loss : 0.58964
Epoch : 9 | Iteration : 186 | Learning Rate : 1.1969e-06 | Total Loss : 238.5719 | Box Loss : 2.0229 | Object Loss : 236.0748 | Class Loss : 0.47414
Epoch : 9 | Iteration : 187 | Learning Rate : 1.2228e-06 | Total Loss : 224.8348 | Box Loss : 1.2722 | Object Loss : 223.0252 | Class Loss : 0.53734
Epoch : 9 | Iteration : 188 | Learning Rate : 1.2492e-06 | Total Loss : 221.2518 | Box Loss : 1.8325 | Object Loss : 218.8065 | Class Loss : 0.61271
Epoch : 9 | Iteration : 189 | Learning Rate : 1.276e-06 | Total Loss : 215.7116 | Box Loss : 1.0737 | Object Loss : 214.1493 | Class Loss : 0.48865
Epoch : 9 | Iteration : 190 | Learning Rate : 1.3032e-06 | Total Loss : 208.2681 | Box Loss : 3.0667 | Object Loss : 204.7304 | Class Loss : 0.47088
Epoch : 9 | Iteration : 191 | Learning Rate : 1.3309e-06 | Total Loss : 204.9479 | Box Loss : 2.8997 | Object Loss : 201.2484 | Class Loss : 0.79992
Epoch : 9 | Iteration : 192 | Learning Rate : 1.359e-06 | Total Loss : 194.3042 | Box Loss : 1.6634 | Object Loss : 192.0485 | Class Loss : 0.59222
Epoch : 9 | Iteration : 193 | Learning Rate : 1.3875e-06 | Total Loss : 192.862 | Box Loss : 2.6475 | Object Loss : 189.624 | Class Loss : 0.59048
Epoch : 9 | Iteration : 194 | Learning Rate : 1.4165e-06 | Total Loss : 185.9765 | Box Loss : 1.4347 | Object Loss : 184.0554 | Class Loss : 0.48638
Epoch : 9 | Iteration : 195 | Learning Rate : 1.4459e-06 | Total Loss : 181.3524 | Box Loss : 2.5645 | Object Loss : 178.0999 | Class Loss : 0.68799
Epoch : 9 | Iteration : 196 | Learning Rate : 1.4758e-06 | Total Loss : 175.0837 | Box Loss : 1.1376 | Object Loss : 173.5423 | Class Loss : 0.40379
Epoch : 9 | Iteration : 197 | Learning Rate : 1.5061e-06 | Total Loss : 172.47 | Box Loss : 2.7052 | Object Loss : 169.3542 | Class Loss : 0.41053
Epoch : 9 | Iteration : 198 | Learning Rate : 1.537e-06 | Total Loss : 167.8906 | Box Loss : 2.1754 | Object Loss : 165.1792 | Class Loss : 0.53594
Epoch : 9 | Iteration : 199 | Learning Rate : 1.5682e-06 | Total Loss : 163.7262 | Box Loss : 3.1509 | Object Loss : 160.1136 | Class Loss : 0.46167
Epoch : 9 | Iteration : 200 | Learning Rate : 1.6e-06 | Total Loss : 157.0047 | Box Loss : 1.8896 | Object Loss : 154.6646 | Class Loss : 0.45044
Epoch : 9 | Iteration : 201 | Learning Rate : 1.6322e-06 | Total Loss : 155.4718 | Box Loss : 2.9187 | Object Loss : 151.9122 | Class Loss : 0.64093
Epoch : 9 | Iteration : 202 | Learning Rate : 1.665e-06 | Total Loss : 150.253 | Box Loss : 1.4516 | Object Loss : 148.2571 | Class Loss : 0.54436
Epoch : 9 | Iteration : 203 | Learning Rate : 1.6982e-06 | Total Loss : 147.4376 | Box Loss : 1.4957 | Object Loss : 145.1635 | Class Loss : 0.77833
Epoch : 9 | Iteration : 204 | Learning Rate : 1.7319e-06 | Total Loss : 143.4381 | Box Loss : 1.4524 | Object Loss : 141.2332 | Class Loss : 0.75252
Epoch : 9 | Iteration : 205 | Learning Rate : 1.7661e-06 | Total Loss : 139.4781 | Box Loss : 2.1622 | Object Loss : 136.841 | Class Loss : 0.47489
Epoch : 9 | Iteration : 206 | Learning Rate : 1.8008e-06 | Total Loss : 135.4959 | Box Loss : 1.7917 | Object Loss : 133.076 | Class Loss : 0.62819
Epoch : 9 | Iteration : 207 | Learning Rate : 1.836e-06 | Total Loss : 151.6894 | Box Loss : 16.8452 | Object Loss : 133.559 | Class Loss : 1.2852
Epoch : 10 | Iteration : 208 | Learning Rate : 1.8718e-06 | Total Loss : 130.3298 | Box Loss : 3.2905 | Object Loss : 126.4972 | Class Loss : 0.54209
Epoch : 10 | Iteration : 209 | Learning Rate : 1.908e-06 | Total Loss : 123.955 | Box Loss : 0.64779 | Object Loss : 122.8501 | Class Loss : 0.45706
Epoch : 10 | Iteration : 210 | Learning Rate : 1.9448e-06 | Total Loss : 122.9245 | Box Loss : 2.1818 | Object Loss : 120.316 | Class Loss : 0.42669
Epoch : 10 | Iteration : 211 | Learning Rate : 1.9821e-06 | Total Loss : 120.8997 | Box Loss : 1.5668 | Object Loss : 118.8077 | Class Loss : 0.52523
Epoch : 10 | Iteration : 212 | Learning Rate : 2.02e-06 | Total Loss : 118.3063 | Box Loss : 2.2124 | Object Loss : 115.522 | Class Loss : 0.57182
Epoch : 10 | Iteration : 213 | Learning Rate : 2.0583e-06 | Total Loss : 113.0302 | Box Loss : 1.4393 | Object Loss : 111.2402 | Class Loss : 0.35071
Epoch : 10 | Iteration : 214 | Learning Rate : 2.0973e-06 | Total Loss : 111.292 | Box Loss : 1.5763 | Object Loss : 109.3507 | Class Loss : 0.36501
Epoch : 10 | Iteration : 215 | Learning Rate : 2.1368e-06 | Total Loss : 110.2187 | Box Loss : 3.697 | Object Loss : 105.9979 | Class Loss : 0.52385
Epoch : 10 | Iteration : 216 | Learning Rate : 2.1768e-06 | Total Loss : 107.9202 | Box Loss : 2.8203 | Object Loss : 104.4144 | Class Loss : 0.68554
Epoch : 10 | Iteration : 217 | Learning Rate : 2.2174e-06 | Total Loss : 105.8905 | Box Loss : 1.6596 | Object Loss : 103.7231 | Class Loss : 0.50786
Epoch : 10 | Iteration : 218 | Learning Rate : 2.2585e-06 | Total Loss : 103.5545 | Box Loss : 2.885 | Object Loss : 99.9706 | Class Loss : 0.6989
Epoch : 10 | Iteration : 219 | Learning Rate : 2.3003e-06 | Total Loss : 97.551 | Box Loss : 0.8408 | Object Loss : 96.1815 | Class Loss : 0.52869
Epoch : 10 | Iteration : 220 | Learning Rate : 2.3426e-06 | Total Loss : 96.6629 | Box Loss : 0.75809 | Object Loss : 95.4954 | Class Loss : 0.40937
Epoch : 10 | Iteration : 221 | Learning Rate : 2.3854e-06 | Total Loss : 96.9953 | Box Loss : 4.1857 | Object Loss : 92.1798 | Class Loss : 0.6298
Epoch : 10 | Iteration : 222 | Learning Rate : 2.4289e-06 | Total Loss : 93.509 | Box Loss : 2.001 | Object Loss : 90.8811 | Class Loss : 0.62687
Epoch : 10 | Iteration : 223 | Learning Rate : 2.473e-06 | Total Loss : 91.9346 | Box Loss : 1.7513 | Object Loss : 89.7099 | Class Loss : 0.47336
Epoch : 10 | Iteration : 224 | Learning Rate : 2.5176e-06 | Total Loss : 88.4763 | Box Loss : 2.1265 | Object Loss : 85.839 | Class Loss : 0.5108
Epoch : 10 | Iteration : 225 | Learning Rate : 2.5629e-06 | Total Loss : 85.629 | Box Loss : 0.98446 | Object Loss : 84.2665 | Class Loss : 0.3781
Epoch : 10 | Iteration : 226 | Learning Rate : 2.6088e-06 | Total Loss : 86.2569 | Box Loss : 2.1277 | Object Loss : 83.5847 | Class Loss : 0.54448
Epoch : 10 | Iteration : 227 | Learning Rate : 2.6552e-06 | Total Loss : 83.1799 | Box Loss : 1.347 | Object Loss : 81.3644 | Class Loss : 0.46843
Epoch : 10 | Iteration : 228 | Learning Rate : 2.7023e-06 | Total Loss : 81.1243 | Box Loss : 1.4545 | Object Loss : 78.8732 | Class Loss : 0.79659
Epoch : 10 | Iteration : 229 | Learning Rate : 2.7501e-06 | Total Loss : 81.2041 | Box Loss : 2.5103 | Object Loss : 78.0889 | Class Loss : 0.60492
Epoch : 10 | Iteration : 230 | Learning Rate : 2.7984e-06 | Total Loss : 75.1246 | Box Loss : 0.27143 | Object Loss : 74.5554 | Class Loss : 0.29782
Epoch : 11 | Iteration : 231 | Learning Rate : 2.8474e-06 | Total Loss : 79.0376 | Box Loss : 3.4272 | Object Loss : 75.0811 | Class Loss : 0.52923
Epoch : 11 | Iteration : 232 | Learning Rate : 2.897e-06 | Total Loss : 73.4143 | Box Loss : 0.50453 | Object Loss : 72.4067 | Class Loss : 0.50303
Epoch : 11 | Iteration : 233 | Learning Rate : 2.9473e-06 | Total Loss : 74.4846 | Box Loss : 1.8336 | Object Loss : 71.9536 | Class Loss : 0.69742
Epoch : 11 | Iteration : 234 | Learning Rate : 2.9982e-06 | Total Loss : 72.1833 | Box Loss : 1.3446 | Object Loss : 70.3828 | Class Loss : 0.45593
Epoch : 11 | Iteration : 235 | Learning Rate : 3.0498e-06 | Total Loss : 70.0606 | Box Loss : 2.1287 | Object Loss : 67.4821 | Class Loss : 0.44973
Epoch : 11 | Iteration : 236 | Learning Rate : 3.102e-06 | Total Loss : 69.5047 | Box Loss : 2.2682 | Object Loss : 66.7052 | Class Loss : 0.53131
Epoch : 11 | Iteration : 237 | Learning Rate : 3.155e-06 | Total Loss : 67.5093 | Box Loss : 1.4033 | Object Loss : 65.7894 | Class Loss : 0.31664
Epoch : 11 | Iteration : 238 | Learning Rate : 3.2085e-06 | Total Loss : 65.0801 | Box Loss : 1.0371 | Object Loss : 63.5287 | Class Loss : 0.51429
Epoch : 11 | Iteration : 239 | Learning Rate : 3.2628e-06 | Total Loss : 65.2496 | Box Loss : 1.5765 | Object Loss : 63.0764 | Class Loss : 0.59679
Epoch : 11 | Iteration : 240 | Learning Rate : 3.3178e-06 | Total Loss : 64.575 | Box Loss : 1.7407 | Object Loss : 62.1388 | Class Loss : 0.69552
Epoch : 11 | Iteration : 241 | Learning Rate : 3.3734e-06 | Total Loss : 64.5196 | Box Loss : 2.1123 | Object Loss : 61.7703 | Class Loss : 0.63696
Epoch : 11 | Iteration : 242 | Learning Rate : 3.4297e-06 | Total Loss : 61.7999 | Box Loss : 2.0481 | Object Loss : 59.1433 | Class Loss : 0.60848
Epoch : 11 | Iteration : 243 | Learning Rate : 3.4868e-06 | Total Loss : 59.071 | Box Loss : 1.6958 | Object Loss : 56.9139 | Class Loss : 0.46125
Epoch : 11 | Iteration : 244 | Learning Rate : 3.5445e-06 | Total Loss : 60.6312 | Box Loss : 2.3208 | Object Loss : 57.7375 | Class Loss : 0.57292
Epoch : 11 | Iteration : 245 | Learning Rate : 3.603e-06 | Total Loss : 56.0652 | Box Loss : 0.81476 | Object Loss : 54.8619 | Class Loss : 0.38854
Epoch : 11 | Iteration : 246 | Learning Rate : 3.6622e-06 | Total Loss : 55.6922 | Box Loss : 0.96189 | Object Loss : 54.3759 | Class Loss : 0.35436
Epoch : 11 | Iteration : 247 | Learning Rate : 3.7221e-06 | Total Loss : 54.9431 | Box Loss : 1.6455 | Object Loss : 52.815 | Class Loss : 0.48265
Epoch : 11 | Iteration : 248 | Learning Rate : 3.7827e-06 | Total Loss : 54.0564 | Box Loss : 1.4288 | Object Loss : 52.3493 | Class Loss : 0.27834
Epoch : 11 | Iteration : 249 | Learning Rate : 3.8441e-06 | Total Loss : 54.5192 | Box Loss : 1.6075 | Object Loss : 52.3839 | Class Loss : 0.52773
Epoch : 11 | Iteration : 250 | Learning Rate : 3.9063e-06 | Total Loss : 50.9148 | Box Loss : 0.64433 | Object Loss : 49.8191 | Class Loss : 0.45141
Epoch : 11 | Iteration : 251 | Learning Rate : 3.9691e-06 | Total Loss : 49.9751 | Box Loss : 1.0272 | Object Loss : 48.508 | Class Loss : 0.4399
Epoch : 11 | Iteration : 252 | Learning Rate : 4.0328e-06 | Total Loss : 51.7999 | Box Loss : 1.8292 | Object Loss : 49.2302 | Class Loss : 0.74052
Epoch : 11 | Iteration : 253 | Learning Rate : 4.0972e-06 | Total Loss : 53.6939 | Box Loss : 3.1018 | Object Loss : 49.5866 | Class Loss : 1.0054
Epoch : 12 | Iteration : 254 | Learning Rate : 4.1623e-06 | Total Loss : 48.7768 | Box Loss : 0.98468 | Object Loss : 47.3193 | Class Loss : 0.47286
Epoch : 12 | Iteration : 255 | Learning Rate : 4.2283e-06 | Total Loss : 49.4233 | Box Loss : 1.9577 | Object Loss : 46.9834 | Class Loss : 0.48223
Epoch : 12 | Iteration : 256 | Learning Rate : 4.295e-06 | Total Loss : 45.1711 | Box Loss : 0.52608 | Object Loss : 44.2151 | Class Loss : 0.42999
Epoch : 12 | Iteration : 257 | Learning Rate : 4.3625e-06 | Total Loss : 47.404 | Box Loss : 1.4689 | Object Loss : 45.4146 | Class Loss : 0.52055
Epoch : 12 | Iteration : 258 | Learning Rate : 4.4308e-06 | Total Loss : 45.8223 | Box Loss : 1.727 | Object Loss : 43.6486 | Class Loss : 0.44666
Epoch : 12 | Iteration : 259 | Learning Rate : 4.4999e-06 | Total Loss : 44.0125 | Box Loss : 1.4235 | Object Loss : 42.2035 | Class Loss : 0.38551
Epoch : 12 | Iteration : 260 | Learning Rate : 4.5698e-06 | Total Loss : 43.6457 | Box Loss : 1.0148 | Object Loss : 42.264 | Class Loss : 0.36687
Epoch : 12 | Iteration : 261 | Learning Rate : 4.6405e-06 | Total Loss : 42.8426 | Box Loss : 1.0523 | Object Loss : 41.2832 | Class Loss : 0.50707
Epoch : 12 | Iteration : 262 | Learning Rate : 4.712e-06 | Total Loss : 42.7025 | Box Loss : 1.6951 | Object Loss : 40.5779 | Class Loss : 0.42952
Epoch : 12 | Iteration : 263 | Learning Rate : 4.7844e-06 | Total Loss : 41.9598 | Box Loss : 1.1832 | Object Loss : 40.2526 | Class Loss : 0.52401
Epoch : 12 | Iteration : 264 | Learning Rate : 4.8575e-06 | Total Loss : 42.3087 | Box Loss : 1.4446 | Object Loss : 40.2068 | Class Loss : 0.65728
Epoch : 12 | Iteration : 265 | Learning Rate : 4.9316e-06 | Total Loss : 42.6846 | Box Loss : 2.1949 | Object Loss : 39.9024 | Class Loss : 0.58729
Epoch : 12 | Iteration : 266 | Learning Rate : 5.0064e-06 | Total Loss : 40.3707 | Box Loss : 1.7943 | Object Loss : 38.0154 | Class Loss : 0.56103
Epoch : 12 | Iteration : 267 | Learning Rate : 5.0821e-06 | Total Loss : 38.4586 | Box Loss : 0.97707 | Object Loss : 37.0044 | Class Loss : 0.4772
Epoch : 12 | Iteration : 268 | Learning Rate : 5.1587e-06 | Total Loss : 38.25 | Box Loss : 0.83939 | Object Loss : 37.0145 | Class Loss : 0.39606
Epoch : 12 | Iteration : 269 | Learning Rate : 5.2361e-06 | Total Loss : 36.7243 | Box Loss : 0.36276 | Object Loss : 35.9324 | Class Loss : 0.42916
Epoch : 12 | Iteration : 270 | Learning Rate : 5.3144e-06 | Total Loss : 36.9852 | Box Loss : 1.5355 | Object Loss : 35.0721 | Class Loss : 0.37763
Epoch : 12 | Iteration : 271 | Learning Rate : 5.3936e-06 | Total Loss : 37.2974 | Box Loss : 1.3626 | Object Loss : 35.5152 | Class Loss : 0.41965
Epoch : 12 | Iteration : 272 | Learning Rate : 5.4736e-06 | Total Loss : 35.8527 | Box Loss : 0.80992 | Object Loss : 34.4731 | Class Loss : 0.56959
Epoch : 12 | Iteration : 273 | Learning Rate : 5.5546e-06 | Total Loss : 37.5231 | Box Loss : 1.6429 | Object Loss : 35.2527 | Class Loss : 0.62757
Epoch : 12 | Iteration : 274 | Learning Rate : 5.6364e-06 | Total Loss : 35.3775 | Box Loss : 1.3198 | Object Loss : 33.6527 | Class Loss : 0.40501
Epoch : 12 | Iteration : 275 | Learning Rate : 5.7191e-06 | Total Loss : 34.1811 | Box Loss : 1.5352 | Object Loss : 32.298 | Class Loss : 0.34784
Epoch : 12 | Iteration : 276 | Learning Rate : 5.8028e-06 | Total Loss : 32.1206 | Box Loss : 0.072668 | Object Loss : 31.8403 | Class Loss : 0.20765
Epoch : 13 | Iteration : 277 | Learning Rate : 5.8873e-06 | Total Loss : 32.9848 | Box Loss : 0.83384 | Object Loss : 31.6276 | Class Loss : 0.52341
Epoch : 13 | Iteration : 278 | Learning Rate : 5.9728e-06 | Total Loss : 33.6007 | Box Loss : 1.1468 | Object Loss : 31.9988 | Class Loss : 0.45515
Epoch : 13 | Iteration : 279 | Learning Rate : 6.0592e-06 | Total Loss : 32.4009 | Box Loss : 1.6988 | Object Loss : 30.308 | Class Loss : 0.39409
Epoch : 13 | Iteration : 280 | Learning Rate : 6.1466e-06 | Total Loss : 32.788 | Box Loss : 1.4912 | Object Loss : 30.8687 | Class Loss : 0.4282
Epoch : 13 | Iteration : 281 | Learning Rate : 6.2348e-06 | Total Loss : 33.6716 | Box Loss : 1.5219 | Object Loss : 31.6115 | Class Loss : 0.53822
Epoch : 13 | Iteration : 282 | Learning Rate : 6.3241e-06 | Total Loss : 31.028 | Box Loss : 1.098 | Object Loss : 29.5333 | Class Loss : 0.3967
Epoch : 13 | Iteration : 283 | Learning Rate : 6.4142e-06 | Total Loss : 30.2471 | Box Loss : 0.81254 | Object Loss : 29.06 | Class Loss : 0.37458
Epoch : 13 | Iteration : 284 | Learning Rate : 6.5054e-06 | Total Loss : 30.5616 | Box Loss : 1.4204 | Object Loss : 28.7683 | Class Loss : 0.37296
Epoch : 13 | Iteration : 285 | Learning Rate : 6.5975e-06 | Total Loss : 29.9051 | Box Loss : 1.154 | Object Loss : 28.4039 | Class Loss : 0.34721
Epoch : 13 | Iteration : 286 | Learning Rate : 6.6906e-06 | Total Loss : 29.5435 | Box Loss : 1.1525 | Object Loss : 27.906 | Class Loss : 0.48489
Epoch : 13 | Iteration : 287 | Learning Rate : 6.7847e-06 | Total Loss : 28.0475 | Box Loss : 0.90402 | Object Loss : 26.8044 | Class Loss : 0.33906
Epoch : 13 | Iteration : 288 | Learning Rate : 6.8797e-06 | Total Loss : 30.9531 | Box Loss : 1.9343 | Object Loss : 28.4668 | Class Loss : 0.55198
Epoch : 13 | Iteration : 289 | Learning Rate : 6.9758e-06 | Total Loss : 29.5509 | Box Loss : 1.6881 | Object Loss : 27.4405 | Class Loss : 0.4222
Epoch : 13 | Iteration : 290 | Learning Rate : 7.0728e-06 | Total Loss : 27.9723 | Box Loss : 1.0452 | Object Loss : 26.4545 | Class Loss : 0.4726
Epoch : 13 | Iteration : 291 | Learning Rate : 7.1709e-06 | Total Loss : 27.0647 | Box Loss : 0.92261 | Object Loss : 25.8066 | Class Loss : 0.33554
Epoch : 13 | Iteration : 292 | Learning Rate : 7.2699e-06 | Total Loss : 26.2642 | Box Loss : 0.7257 | Object Loss : 25.2021 | Class Loss : 0.33634
Epoch : 13 | Iteration : 293 | Learning Rate : 7.3701e-06 | Total Loss : 26.0426 | Box Loss : 0.84798 | Object Loss : 24.8537 | Class Loss : 0.3409
Epoch : 13 | Iteration : 294 | Learning Rate : 7.4712e-06 | Total Loss : 25.1769 | Box Loss : 0.34435 | Object Loss : 24.4557 | Class Loss : 0.37682
Epoch : 13 | Iteration : 295 | Learning Rate : 7.5734e-06 | Total Loss : 25.8779 | Box Loss : 0.73083 | Object Loss : 24.5958 | Class Loss : 0.55121
Epoch : 13 | Iteration : 296 | Learning Rate : 7.6766e-06 | Total Loss : 25.0045 | Box Loss : 0.76468 | Object Loss : 23.8644 | Class Loss : 0.37537
Epoch : 13 | Iteration : 297 | Learning Rate : 7.7808e-06 | Total Loss : 27.1041 | Box Loss : 1.8114 | Object Loss : 24.7396 | Class Loss : 0.55318
Epoch : 13 | Iteration : 298 | Learning Rate : 7.8862e-06 | Total Loss : 27.4797 | Box Loss : 1.6207 | Object Loss : 25.301 | Class Loss : 0.55801
Epoch : 13 | Iteration : 299 | Learning Rate : 7.9925e-06 | Total Loss : 26.4747 | Box Loss : 0.94607 | Object Loss : 25.2766 | Class Loss : 0.25199
Epoch : 14 | Iteration : 300 | Learning Rate : 8.1e-06 | Total Loss : 23.4735 | Box Loss : 0.65068 | Object Loss : 22.5667 | Class Loss : 0.2561
Epoch : 14 | Iteration : 301 | Learning Rate : 8.2085e-06 | Total Loss : 25.7068 | Box Loss : 1.0861 | Object Loss : 24.149 | Class Loss : 0.47174
Epoch : 14 | Iteration : 302 | Learning Rate : 8.3182e-06 | Total Loss : 23.1165 | Box Loss : 0.90589 | Object Loss : 21.9181 | Class Loss : 0.29255
Epoch : 14 | Iteration : 303 | Learning Rate : 8.4289e-06 | Total Loss : 22.5633 | Box Loss : 0.73095 | Object Loss : 21.4559 | Class Loss : 0.37652
Epoch : 14 | Iteration : 304 | Learning Rate : 8.5407e-06 | Total Loss : 24.2982 | Box Loss : 1.4886 | Object Loss : 22.3874 | Class Loss : 0.4222
Epoch : 14 | Iteration : 305 | Learning Rate : 8.6537e-06 | Total Loss : 22.2776 | Box Loss : 0.94406 | Object Loss : 20.998 | Class Loss : 0.33558
Epoch : 14 | Iteration : 306 | Learning Rate : 8.7677e-06 | Total Loss : 22.1669 | Box Loss : 1.0104 | Object Loss : 20.851 | Class Loss : 0.30545
Epoch : 14 | Iteration : 307 | Learning Rate : 8.8829e-06 | Total Loss : 23.0523 | Box Loss : 1.2233 | Object Loss : 21.1282 | Class Loss : 0.70076
Epoch : 14 | Iteration : 308 | Learning Rate : 8.9992e-06 | Total Loss : 22.5736 | Box Loss : 0.86523 | Object Loss : 21.3164 | Class Loss : 0.392
Epoch : 14 | Iteration : 309 | Learning Rate : 9.1166e-06 | Total Loss : 23.4343 | Box Loss : 1.4781 | Object Loss : 21.4404 | Class Loss : 0.51575
Epoch : 14 | Iteration : 310 | Learning Rate : 9.2352e-06 | Total Loss : 22.5019 | Box Loss : 0.90839 | Object Loss : 21.13 | Class Loss : 0.46357
Epoch : 14 | Iteration : 311 | Learning Rate : 9.355e-06 | Total Loss : 22.6707 | Box Loss : 1.0287 | Object Loss : 21.2311 | Class Loss : 0.41088
Epoch : 14 | Iteration : 312 | Learning Rate : 9.4759e-06 | Total Loss : 20.3037 | Box Loss : 0.7972 | Object Loss : 19.2077 | Class Loss : 0.29872
Epoch : 14 | Iteration : 313 | Learning Rate : 9.5979e-06 | Total Loss : 24.9406 | Box Loss : 2.2247 | Object Loss : 22.0054 | Class Loss : 0.7105
Epoch : 14 | Iteration : 314 | Learning Rate : 9.7212e-06 | Total Loss : 21.6491 | Box Loss : 1.3531 | Object Loss : 19.9613 | Class Loss : 0.33469
Epoch : 14 | Iteration : 315 | Learning Rate : 9.8456e-06 | Total Loss : 19.9248 | Box Loss : 0.77617 | Object Loss : 18.7892 | Class Loss : 0.35945
Epoch : 14 | Iteration : 316 | Learning Rate : 9.9712e-06 | Total Loss : 18.9517 | Box Loss : 0.53695 | Object Loss : 18.188 | Class Loss : 0.22678
Epoch : 14 | Iteration : 317 | Learning Rate : 1.0098e-05 | Total Loss : 22.1042 | Box Loss : 1.7879 | Object Loss : 19.8364 | Class Loss : 0.47995
Epoch : 14 | Iteration : 318 | Learning Rate : 1.0226e-05 | Total Loss : 18.7485 | Box Loss : 0.62062 | Object Loss : 17.7331 | Class Loss : 0.39474
Epoch : 14 | Iteration : 319 | Learning Rate : 1.0355e-05 | Total Loss : 20.3975 | Box Loss : 0.87357 | Object Loss : 19.0656 | Class Loss : 0.45827
Epoch : 14 | Iteration : 320 | Learning Rate : 1.0486e-05 | Total Loss : 19.8063 | Box Loss : 1.4676 | Object Loss : 17.9057 | Class Loss : 0.43309
Epoch : 14 | Iteration : 321 | Learning Rate : 1.0617e-05 | Total Loss : 19.4166 | Box Loss : 0.75374 | Object Loss : 18.2976 | Class Loss : 0.36532
Epoch : 14 | Iteration : 322 | Learning Rate : 1.075e-05 | Total Loss : 24.5432 | Box Loss : 1.4762 | Object Loss : 22.4664 | Class Loss : 0.60069
Epoch : 15 | Iteration : 323 | Learning Rate : 1.0885e-05 | Total Loss : 18.4276 | Box Loss : 0.98014 | Object Loss : 17.1027 | Class Loss : 0.34474
Epoch : 15 | Iteration : 324 | Learning Rate : 1.102e-05 | Total Loss : 20.2738 | Box Loss : 1.2267 | Object Loss : 18.6558 | Class Loss : 0.39124
Epoch : 15 | Iteration : 325 | Learning Rate : 1.1157e-05 | Total Loss : 17.5261 | Box Loss : 0.7839 | Object Loss : 16.4204 | Class Loss : 0.32173
Epoch : 15 | Iteration : 326 | Learning Rate : 1.1295e-05 | Total Loss : 20.0305 | Box Loss : 1.0768 | Object Loss : 18.4377 | Class Loss : 0.51609
Epoch : 15 | Iteration : 327 | Learning Rate : 1.1434e-05 | Total Loss : 18.1276 | Box Loss : 1.084 | Object Loss : 16.5841 | Class Loss : 0.45952
Epoch : 15 | Iteration : 328 | Learning Rate : 1.1574e-05 | Total Loss : 16.9899 | Box Loss : 0.52159 | Object Loss : 16.25 | Class Loss : 0.21836
Epoch : 15 | Iteration : 329 | Learning Rate : 1.1716e-05 | Total Loss : 17.2788 | Box Loss : 0.79391 | Object Loss : 16.2221 | Class Loss : 0.26281
Epoch : 15 | Iteration : 330 | Learning Rate : 1.1859e-05 | Total Loss : 16.7677 | Box Loss : 0.59361 | Object Loss : 15.8916 | Class Loss : 0.28244
Epoch : 15 | Iteration : 331 | Learning Rate : 1.2004e-05 | Total Loss : 17.8247 | Box Loss : 0.71991 | Object Loss : 16.5339 | Class Loss : 0.57095
Epoch : 15 | Iteration : 332 | Learning Rate : 1.2149e-05 | Total Loss : 17.9062 | Box Loss : 0.95895 | Object Loss : 16.52 | Class Loss : 0.42724
Epoch : 15 | Iteration : 333 | Learning Rate : 1.2296e-05 | Total Loss : 17.3879 | Box Loss : 0.81765 | Object Loss : 16.1917 | Class Loss : 0.37862
Epoch : 15 | Iteration : 334 | Learning Rate : 1.2445e-05 | Total Loss : 15.6942 | Box Loss : 0.55038 | Object Loss : 14.9074 | Class Loss : 0.23641
Epoch : 15 | Iteration : 335 | Learning Rate : 1.2594e-05 | Total Loss : 15.7281 | Box Loss : 0.60656 | Object Loss : 14.9006 | Class Loss : 0.22098
Epoch : 15 | Iteration : 336 | Learning Rate : 1.2746e-05 | Total Loss : 16.8435 | Box Loss : 0.81891 | Object Loss : 15.6955 | Class Loss : 0.32907
Epoch : 15 | Iteration : 337 | Learning Rate : 1.2898e-05 | Total Loss : 20.7239 | Box Loss : 1.6438 | Object Loss : 18.4175 | Class Loss : 0.66266
Epoch : 15 | Iteration : 338 | Learning Rate : 1.3052e-05 | Total Loss : 15.5238 | Box Loss : 0.49358 | Object Loss : 14.7686 | Class Loss : 0.2616
Epoch : 15 | Iteration : 339 | Learning Rate : 1.3207e-05 | Total Loss : 16.3684 | Box Loss : 0.87811 | Object Loss : 15.1227 | Class Loss : 0.36764
Epoch : 15 | Iteration : 340 | Learning Rate : 1.3363e-05 | Total Loss : 15.784 | Box Loss : 0.90518 | Object Loss : 14.5701 | Class Loss : 0.30873
Epoch : 15 | Iteration : 341 | Learning Rate : 1.3521e-05 | Total Loss : 18.1031 | Box Loss : 1.9281 | Object Loss : 15.805 | Class Loss : 0.36997
Epoch : 15 | Iteration : 342 | Learning Rate : 1.3681e-05 | Total Loss : 17.0571 | Box Loss : 1.5809 | Object Loss : 15.1142 | Class Loss : 0.36199
Epoch : 15 | Iteration : 343 | Learning Rate : 1.3841e-05 | Total Loss : 16.2661 | Box Loss : 0.96523 | Object Loss : 14.9404 | Class Loss : 0.36045
Epoch : 15 | Iteration : 344 | Learning Rate : 1.4003e-05 | Total Loss : 15.5492 | Box Loss : 0.8949 | Object Loss : 14.3923 | Class Loss : 0.26203
Epoch : 15 | Iteration : 345 | Learning Rate : 1.4167e-05 | Total Loss : 20.312 | Box Loss : 1.527 | Object Loss : 18.1565 | Class Loss : 0.62849
Epoch : 16 | Iteration : 346 | Learning Rate : 1.4332e-05 | Total Loss : 17.2633 | Box Loss : 1.634 | Object Loss : 15.2527 | Class Loss : 0.37669
Epoch : 16 | Iteration : 347 | Learning Rate : 1.4498e-05 | Total Loss : 14.8747 | Box Loss : 0.7295 | Object Loss : 13.7303 | Class Loss : 0.41487
Epoch : 16 | Iteration : 348 | Learning Rate : 1.4666e-05 | Total Loss : 14.2806 | Box Loss : 0.53956 | Object Loss : 13.4684 | Class Loss : 0.27264
Epoch : 16 | Iteration : 349 | Learning Rate : 1.4835e-05 | Total Loss : 16.0981 | Box Loss : 1.0712 | Object Loss : 14.523 | Class Loss : 0.50388
Epoch : 16 | Iteration : 350 | Learning Rate : 1.5006e-05 | Total Loss : 13.002 | Box Loss : 0.43276 | Object Loss : 12.3231 | Class Loss : 0.24609
Epoch : 16 | Iteration : 351 | Learning Rate : 1.5178e-05 | Total Loss : 14.86 | Box Loss : 1.3465 | Object Loss : 13.2465 | Class Loss : 0.26695
Epoch : 16 | Iteration : 352 | Learning Rate : 1.5352e-05 | Total Loss : 13.5714 | Box Loss : 0.93982 | Object Loss : 12.4149 | Class Loss : 0.21666
Epoch : 16 | Iteration : 353 | Learning Rate : 1.5527e-05 | Total Loss : 14.0531 | Box Loss : 0.75712 | Object Loss : 12.9685 | Class Loss : 0.32748
Epoch : 16 | Iteration : 354 | Learning Rate : 1.5704e-05 | Total Loss : 12.627 | Box Loss : 0.40763 | Object Loss : 12.0022 | Class Loss : 0.21726
Epoch : 16 | Iteration : 355 | Learning Rate : 1.5882e-05 | Total Loss : 13.1038 | Box Loss : 0.32031 | Object Loss : 12.563 | Class Loss : 0.2205
Epoch : 16 | Iteration : 356 | Learning Rate : 1.6062e-05 | Total Loss : 15.1351 | Box Loss : 1.3416 | Object Loss : 13.4434 | Class Loss : 0.35006
Epoch : 16 | Iteration : 357 | Learning Rate : 1.6243e-05 | Total Loss : 13.5998 | Box Loss : 0.86842 | Object Loss : 12.5723 | Class Loss : 0.15913
Epoch : 16 | Iteration : 358 | Learning Rate : 1.6426e-05 | Total Loss : 17.9834 | Box Loss : 2.2705 | Object Loss : 15.0911 | Class Loss : 0.62189
Epoch : 16 | Iteration : 359 | Learning Rate : 1.661e-05 | Total Loss : 13.2651 | Box Loss : 0.77707 | Object Loss : 12.1661 | Class Loss : 0.32198
Epoch : 16 | Iteration : 360 | Learning Rate : 1.6796e-05 | Total Loss : 12.363 | Box Loss : 0.56516 | Object Loss : 11.603 | Class Loss : 0.1949
Epoch : 16 | Iteration : 361 | Learning Rate : 1.6984e-05 | Total Loss : 16.1314 | Box Loss : 1.521 | Object Loss : 14.0555 | Class Loss : 0.55489
Epoch : 16 | Iteration : 362 | Learning Rate : 1.7173e-05 | Total Loss : 12.5865 | Box Loss : 0.60568 | Object Loss : 11.7713 | Class Loss : 0.20958
Epoch : 16 | Iteration : 363 | Learning Rate : 1.7363e-05 | Total Loss : 15.2939 | Box Loss : 0.98848 | Object Loss : 13.8377 | Class Loss : 0.46774
Epoch : 16 | Iteration : 364 | Learning Rate : 1.7555e-05 | Total Loss : 14.0186 | Box Loss : 1.3655 | Object Loss : 12.2908 | Class Loss : 0.3623
Epoch : 16 | Iteration : 365 | Learning Rate : 1.7749e-05 | Total Loss : 11.411 | Box Loss : 0.53581 | Object Loss : 10.6625 | Class Loss : 0.21268
Epoch : 16 | Iteration : 366 | Learning Rate : 1.7944e-05 | Total Loss : 16.9061 | Box Loss : 2.0613 | Object Loss : 14.1356 | Class Loss : 0.70913
Epoch : 16 | Iteration : 367 | Learning Rate : 1.8141e-05 | Total Loss : 12.6763 | Box Loss : 0.65758 | Object Loss : 11.7863 | Class Loss : 0.23245
Epoch : 16 | Iteration : 368 | Learning Rate : 1.834e-05 | Total Loss : 10.0711 | Box Loss : 0.1887 | Object Loss : 9.6617 | Class Loss : 0.22069
Epoch : 17 | Iteration : 369 | Learning Rate : 1.854e-05 | Total Loss : 12.3867 | Box Loss : 0.68877 | Object Loss : 11.4522 | Class Loss : 0.24571
Epoch : 17 | Iteration : 370 | Learning Rate : 1.8742e-05 | Total Loss : 15.6693 | Box Loss : 1.4483 | Object Loss : 13.7836 | Class Loss : 0.43749
Epoch : 17 | Iteration : 371 | Learning Rate : 1.8945e-05 | Total Loss : 11.2525 | Box Loss : 0.38393 | Object Loss : 10.5911 | Class Loss : 0.27744
Epoch : 17 | Iteration : 372 | Learning Rate : 1.915e-05 | Total Loss : 12.624 | Box Loss : 1.0968 | Object Loss : 11.2567 | Class Loss : 0.2705
Epoch : 17 | Iteration : 373 | Learning Rate : 1.9357e-05 | Total Loss : 14.1601 | Box Loss : 1.9424 | Object Loss : 11.6739 | Class Loss : 0.54376
Epoch : 17 | Iteration : 374 | Learning Rate : 1.9565e-05 | Total Loss : 13.8942 | Box Loss : 1.4335 | Object Loss : 12.0368 | Class Loss : 0.42392
Epoch : 17 | Iteration : 375 | Learning Rate : 1.9775e-05 | Total Loss : 11.7365 | Box Loss : 0.51757 | Object Loss : 10.9588 | Class Loss : 0.2601
Epoch : 17 | Iteration : 376 | Learning Rate : 1.9987e-05 | Total Loss : 13.3418 | Box Loss : 1.1792 | Object Loss : 11.7613 | Class Loss : 0.40132
Epoch : 17 | Iteration : 377 | Learning Rate : 2.0201e-05 | Total Loss : 12.5316 | Box Loss : 0.5521 | Object Loss : 11.662 | Class Loss : 0.31745
Epoch : 17 | Iteration : 378 | Learning Rate : 2.0416e-05 | Total Loss : 11.6554 | Box Loss : 0.71192 | Object Loss : 10.7293 | Class Loss : 0.21421
Epoch : 17 | Iteration : 379 | Learning Rate : 2.0633e-05 | Total Loss : 10.5197 | Box Loss : 0.45873 | Object Loss : 9.8577 | Class Loss : 0.20335
Epoch : 17 | Iteration : 380 | Learning Rate : 2.0851e-05 | Total Loss : 10.8727 | Box Loss : 0.6144 | Object Loss : 10.0022 | Class Loss : 0.25605
Epoch : 17 | Iteration : 381 | Learning Rate : 2.1072e-05 | Total Loss : 12.827 | Box Loss : 1.5994 | Object Loss : 10.7265 | Class Loss : 0.50119
Epoch : 17 | Iteration : 382 | Learning Rate : 2.1294e-05 | Total Loss : 13.5798 | Box Loss : 1.0056 | Object Loss : 12.0783 | Class Loss : 0.49589
Epoch : 17 | Iteration : 383 | Learning Rate : 2.1518e-05 | Total Loss : 10.2467 | Box Loss : 0.38574 | Object Loss : 9.653 | Class Loss : 0.20797
Epoch : 17 | Iteration : 384 | Learning Rate : 2.1743e-05 | Total Loss : 9.5454 | Box Loss : 0.31727 | Object Loss : 9.0364 | Class Loss : 0.19168
Epoch : 17 | Iteration : 385 | Learning Rate : 2.1971e-05 | Total Loss : 10.7994 | Box Loss : 1.0524 | Object Loss : 9.5408 | Class Loss : 0.20626
Epoch : 17 | Iteration : 386 | Learning Rate : 2.22e-05 | Total Loss : 13.8463 | Box Loss : 1.1893 | Object Loss : 12.236 | Class Loss : 0.42102
Epoch : 17 | Iteration : 387 | Learning Rate : 2.2431e-05 | Total Loss : 9.5131 | Box Loss : 0.33435 | Object Loss : 8.9877 | Class Loss : 0.19104
Epoch : 17 | Iteration : 388 | Learning Rate : 2.2663e-05 | Total Loss : 11.7341 | Box Loss : 0.82945 | Object Loss : 10.6175 | Class Loss : 0.28709
Epoch : 17 | Iteration : 389 | Learning Rate : 2.2898e-05 | Total Loss : 12.5049 | Box Loss : 1.1493 | Object Loss : 10.9918 | Class Loss : 0.3639
Epoch : 17 | Iteration : 390 | Learning Rate : 2.3134e-05 | Total Loss : 10.0721 | Box Loss : 0.78613 | Object Loss : 9.0417 | Class Loss : 0.24434
Epoch : 17 | Iteration : 391 | Learning Rate : 2.3373e-05 | Total Loss : 12.3075 | Box Loss : 1.3411 | Object Loss : 10.7143 | Class Loss : 0.25203
Epoch : 18 | Iteration : 392 | Learning Rate : 2.3613e-05 | Total Loss : 9.7277 | Box Loss : 0.45647 | Object Loss : 8.966 | Class Loss : 0.30527
Epoch : 18 | Iteration : 393 | Learning Rate : 2.3854e-05 | Total Loss : 13.1086 | Box Loss : 1.666 | Object Loss : 11.1439 | Class Loss : 0.2987
Epoch : 18 | Iteration : 394 | Learning Rate : 2.4098e-05 | Total Loss : 12.9978 | Box Loss : 0.88292 | Object Loss : 11.6079 | Class Loss : 0.50705
Epoch : 18 | Iteration : 395 | Learning Rate : 2.4344e-05 | Total Loss : 10.7919 | Box Loss : 0.53646 | Object Loss : 9.9745 | Class Loss : 0.28095
Epoch : 18 | Iteration : 396 | Learning Rate : 2.4591e-05 | Total Loss : 9.7264 | Box Loss : 0.7848 | Object Loss : 8.6938 | Class Loss : 0.24779
Epoch : 18 | Iteration : 397 | Learning Rate : 2.4841e-05 | Total Loss : 10.4778 | Box Loss : 1.0495 | Object Loss : 9.2081 | Class Loss : 0.22026
Epoch : 18 | Iteration : 398 | Learning Rate : 2.5092e-05 | Total Loss : 9.169 | Box Loss : 0.57491 | Object Loss : 8.3962 | Class Loss : 0.1979
Epoch : 18 | Iteration : 399 | Learning Rate : 2.5345e-05 | Total Loss : 11.1855 | Box Loss : 1.0002 | Object Loss : 9.9185 | Class Loss : 0.26684
Epoch : 18 | Iteration : 400 | Learning Rate : 2.56e-05 | Total Loss : 10.7894 | Box Loss : 1.1357 | Object Loss : 9.1997 | Class Loss : 0.45402
Epoch : 18 | Iteration : 401 | Learning Rate : 2.5857e-05 | Total Loss : 9.2411 | Box Loss : 0.46813 | Object Loss : 8.5093 | Class Loss : 0.2637
Epoch : 18 | Iteration : 402 | Learning Rate : 2.6116e-05 | Total Loss : 10.2135 | Box Loss : 0.60753 | Object Loss : 9.2861 | Class Loss : 0.31993
Epoch : 18 | Iteration : 403 | Learning Rate : 2.6377e-05 | Total Loss : 11.366 | Box Loss : 0.78865 | Object Loss : 10.2008 | Class Loss : 0.37651
Epoch : 18 | Iteration : 404 | Learning Rate : 2.6639e-05 | Total Loss : 8.2499 | Box Loss : 0.39101 | Object Loss : 7.6322 | Class Loss : 0.22674
Epoch : 18 | Iteration : 405 | Learning ...

Evaluate Model

Computer Vision System Toolbox™ provides object detector evaluation functions to measure common metrics such as average precision (evaluateDetectionPrecision) and log-average miss rates (evaluateDetectionMissRate). In this example, the average precision metric is used. The average precision provides a single number that incorporates the ability of the detector to make correct classifications (precision) and the ability of the detector to find all relevant objects (recall).

results = detect(yolov3Detector,testData,'MiniBatchSize',8);

% Evaluate the object detector using Average Precision metric.
[ap,recall,precision] = evaluateDetectionPrecision(results,testData);

The precision-recall (PR) curve shows how precise a detector is at varying levels of recall. Ideally, the precision is 1 at all recall levels.

% Plot precision-recall curve.
figure
plot(recall,precision)
xlabel('Recall')
ylabel('Precision')
grid on
title(sprintf('Average Precision = %.2f', ap))

Detect Objects Using YOLO v3

Use the detector for object detection.

% Read the datastore.
data = read(testData);

% Get the image.
I = data{1};

[bboxes,scores,labels] = detect(yolov3Detector,I);

% Display the detections on image.
I = insertObjectAnnotation(I,'rectangle',bboxes,scores);

figure
imshow(I)

Supporting Functions

Model Gradients Function

The function modelGradients takes as input the yolov3ObjectDetector object, a mini-batch of input data XTrain with corresponding ground truth boxes YTrain, the specified penalty threshold as input arguments and returns the gradients of the loss with respect to the learnable parameters in yolov3ObjectDetector, the corresponding mini-batch loss information, and the state of the current batch.

The model gradients function computes the total loss and gradients by performing these operations.

• Generate predictions from the input batch of images using the forward method.

• Collect predictions on the CPU for postprocessing.

• Convert the predictions from the YOLO v3 grid cell coordinates to bounding box coordinates to allow easy comparison with the ground truth data by using the anchorBoxGenerator method of yolov3ObjectDetector.

• Generate targets for loss computation by using the converted predictions and ground truth data. These targets are generated for bounding box positions (x, y, width, height), object confidence, and class probabilities. See the supporting function generateTargets.

• Calculates the mean squared error of the predicted bounding box coordinates with target boxes. See the supporting function bboxOffsetLoss.

• Determines the binary cross-entropy of the predicted object confidence score with target object confidence score. See the supporting function objectnessLoss.

• Determines the binary cross-entropy of the predicted class of object with the target. See the supporting function classConfidenceLoss.

• Computes the total loss as the sum of all losses.

• Computes the gradients of learnables with respect to the total loss.

function [gradients, state, info] = modelGradients(detector, XTrain, YTrain, penaltyThreshold)
inputImageSize = size(XTrain,1:2);

% Gather the ground truths in the CPU for post processing
YTrain = gather(extractdata(YTrain));

% Extract the predictions from the detector.
[gatheredPredictions, YPredCell, state] = forward(detector, XTrain);

% Generate target for predictions from the ground truth data.
[boxTarget, objectnessTarget, classTarget, objectMaskTarget, boxErrorScale] = generateTargets(gatheredPredictions,...
    YTrain, inputImageSize, detector.AnchorBoxes, penaltyThreshold);

% Compute the loss.
boxLoss = bboxOffsetLoss(YPredCell(:,[2 3 7 8]),boxTarget,objectMaskTarget,boxErrorScale);
objLoss = objectnessLoss(YPredCell(:,1),objectnessTarget,objectMaskTarget);
clsLoss = classConfidenceLoss(YPredCell(:,6),classTarget,objectMaskTarget);
totalLoss = boxLoss + objLoss + clsLoss;

info.boxLoss = boxLoss;
info.objLoss = objLoss;
info.clsLoss = clsLoss;
info.totalLoss = totalLoss;

% Compute gradients of learnables with regard to loss.
gradients = dlgradient(totalLoss, detector.Learnables);
end

function boxLoss = bboxOffsetLoss(boxPredCell, boxDeltaTarget, boxMaskTarget, boxErrorScaleTarget)
% Mean squared error for bounding box position.
lossX = sum(cellfun(@(a,b,c,d) mse(a.*c.*d,b.*c.*d),boxPredCell(:,1),boxDeltaTarget(:,1),boxMaskTarget(:,1),boxErrorScaleTarget));
lossY = sum(cellfun(@(a,b,c,d) mse(a.*c.*d,b.*c.*d),boxPredCell(:,2),boxDeltaTarget(:,2),boxMaskTarget(:,1),boxErrorScaleTarget));
lossW = sum(cellfun(@(a,b,c,d) mse(a.*c.*d,b.*c.*d),boxPredCell(:,3),boxDeltaTarget(:,3),boxMaskTarget(:,1),boxErrorScaleTarget));
lossH = sum(cellfun(@(a,b,c,d) mse(a.*c.*d,b.*c.*d),boxPredCell(:,4),boxDeltaTarget(:,4),boxMaskTarget(:,1),boxErrorScaleTarget));
boxLoss = lossX+lossY+lossW+lossH;
end

function objLoss = objectnessLoss(objectnessPredCell, objectnessDeltaTarget, boxMaskTarget)
% Binary cross-entropy loss for objectness score.
objLoss = sum(cellfun(@(a,b,c) crossentropy(a.*c,b.*c,'TargetCategories','independent'),objectnessPredCell,objectnessDeltaTarget,boxMaskTarget(:,2)));
end

function clsLoss = classConfidenceLoss(classPredCell, classTarget, boxMaskTarget)
% Binary cross-entropy loss for class confidence score.
clsLoss = sum(cellfun(@(a,b,c) crossentropy(a.*c,b.*c,'TargetCategories','independent'),classPredCell,classTarget,boxMaskTarget(:,3)));
end

Augmentation and Data Processing Functions

function data = augmentData(A)
% Apply random horizontal flipping, and random X/Y scaling. Boxes that get
% scaled outside the bounds are clipped if the overlap is above 0.25. Also,
% jitter image color.

data = cell(size(A));
for ii = 1:size(A,1)
    I = A{ii,1};
    bboxes = A{ii,2};
    labels = A{ii,3};
    sz = size(I);

    if numel(sz) == 3 && sz(3) == 3
        I = jitterColorHSV(I,...
            'Contrast',0.0,...
            'Hue',0.1,...
            'Saturation',0.2,...
            'Brightness',0.2);
    end
    
    % Randomly flip image.
    tform = randomAffine2d('XReflection',true,'Scale',[1 1.1]);
    rout = affineOutputView(sz,tform,'BoundsStyle','centerOutput');
    I = imwarp(I,tform,'OutputView',rout);
    
    % Apply same transform to boxes.
    [bboxes,indices] = bboxwarp(bboxes,tform,rout,'OverlapThreshold',0.25);
    labels = labels(indices);
    
    % Return original data only when all boxes are removed by warping.
    if isempty(indices)
        data(ii,:) = A(ii,:);
    else
        data(ii,:) = {I, bboxes, labels};
    end
end
end


function data = preprocessData(data, targetSize)
% Resize the images and scale the pixels to between 0 and 1. Also scale the
% corresponding bounding boxes.

for ii = 1:size(data,1)
    I = data{ii,1};
    imgSize = size(I);
    
    % Convert an input image with single channel to 3 channels.
    if numel(imgSize) < 3 
        I = repmat(I,1,1,3);
    end
    bboxes = data{ii,2};

    I = im2single(imresize(I,targetSize(1:2)));
    scale = targetSize(1:2)./imgSize(1:2);
    bboxes = bboxresize(bboxes,scale);
    
    data(ii, 1:2) = {I, bboxes};
end
end

function [XTrain, YTrain] = createBatchData(data, groundTruthBoxes, groundTruthClasses, classNames)
% Returns images combined along the batch dimension in XTrain and
% normalized bounding boxes concatenated with classIDs in YTrain

% Concatenate images along the batch dimension.
XTrain = cat(4, data{:,1});

% Get class IDs from the class names.
classNames = repmat({categorical(classNames')}, size(groundTruthClasses));
[~, classIndices] = cellfun(@(a,b)ismember(a,b), groundTruthClasses, classNames, 'UniformOutput', false);

% Append the label indexes and training image size to scaled bounding boxes
% and create a single cell array of responses.
combinedResponses = cellfun(@(bbox, classid)[bbox, classid], groundTruthBoxes, classIndices, 'UniformOutput', false);
len = max( cellfun(@(x)size(x,1), combinedResponses ) );
paddedBBoxes = cellfun( @(v) padarray(v,[len-size(v,1),0],0,'post'), combinedResponses, 'UniformOutput',false);
YTrain = cat(4, paddedBBoxes{:,1});
end

Learning Rate Schedule Function

function currentLR = piecewiseLearningRateWithWarmup(iteration, epoch, learningRate, warmupPeriod, numEpochs)
% The piecewiseLearningRateWithWarmup function computes the current
% learning rate based on the iteration number.
persistent warmUpEpoch;

if iteration <= warmupPeriod
    % Increase the learning rate for number of iterations in warmup period.
    currentLR = learningRate * ((iteration/warmupPeriod)^4);
    warmUpEpoch = epoch;
elseif iteration >= warmupPeriod && epoch < warmUpEpoch+floor(0.6*(numEpochs-warmUpEpoch))
    % After warm up period, keep the learning rate constant if the remaining number of epochs is less than 60 percent. 
    currentLR = learningRate;
    
elseif epoch >= warmUpEpoch + floor(0.6*(numEpochs-warmUpEpoch)) && epoch < warmUpEpoch+floor(0.9*(numEpochs-warmUpEpoch))
    % If the remaining number of epochs is more than 60 percent but less
    % than 90 percent multiply the learning rate by 0.1.
    currentLR = learningRate*0.1;
    
else
    % If remaining epochs are more than 90 percent multiply the learning
    % rate by 0.01.
    currentLR = learningRate*0.01;
end

end

Utility Functions

function [lossPlotter, learningRatePlotter] = configureTrainingProgressPlotter(f)
% Create the subplots to display the loss and learning rate.
figure(f);
clf
subplot(2,1,1);
ylabel('Learning Rate');
xlabel('Iteration');
learningRatePlotter = animatedline;
subplot(2,1,2);
ylabel('Total Loss');
xlabel('Iteration');
lossPlotter = animatedline;
end

function displayLossInfo(epoch, iteration, currentLR, lossInfo)
% Display loss information for each iteration.
disp("Epoch : " + epoch + " | Iteration : " + iteration + " | Learning Rate : " + currentLR + ...
   " | Total Loss : " + double(gather(extractdata(lossInfo.totalLoss))) + ...
   " | Box Loss : " + double(gather(extractdata(lossInfo.boxLoss))) + ...
   " | Object Loss : " + double(gather(extractdata(lossInfo.objLoss))) + ...
   " | Class Loss : " + double(gather(extractdata(lossInfo.clsLoss))));
end

function updatePlots(lossPlotter, learningRatePlotter, iteration, currentLR, totalLoss)
% Update loss and learning rate plots.
addpoints(lossPlotter, iteration, double(extractdata(gather(totalLoss))));
addpoints(learningRatePlotter, iteration, currentLR);
drawnow
end

function detector = downloadPretrainedYOLOv3Detector()
% Download a pretrained yolov3 detector.
if ~exist('yolov3SqueezeNetVehicleExample_21aSPKG.mat', 'file')
    if ~exist('yolov3SqueezeNetVehicleExample_21aSPKG.zip', 'file')
        disp('Downloading pretrained detector...');
        pretrainedURL = 'https://ssd.mathworks.com/supportfiles/vision/data/yolov3SqueezeNetVehicleExample_21aSPKG.zip';
        websave('yolov3SqueezeNetVehicleExample_21aSPKG.zip', pretrainedURL);
    end
    unzip('yolov3SqueezeNetVehicleExample_21aSPKG.zip');
end
pretrained = load("yolov3SqueezeNetVehicleExample_21aSPKG.mat");
detector = pretrained.detector;
end

References

[1] Redmon, Joseph, and Ali Farhadi. “YOLOv3: An Incremental Improvement.” Preprint, submitted April 8, 2018. https://arxiv.org/abs/1804.02767.