- Data augmentation: Data augmentation is particularly useful when you have a limited amount of data. It helps create additional training samples by applying random transformations to your existing images. This can help the model generalize better and improve its performance. MATLAB provides built-in functions for data augmentation. Please refer to the following page for more information: https://www.mathworks.com/help/deeplearning/ref/imagedataaugmenter.html
- Validation set: In addition to splitting your data into training and test sets, it's recommended to have a validation set. The validation set is used to monitor the model's performance during training and tune hyperparameters. This will help you avoid overfitting and select the best model.
- Object detection evaluation: When evaluating your Faster R-CNN model for object detection, it's important to use appropriate evaluation metrics such as mean Average Precision (mAP) to measure the model's accuracy. MATLAB provides functions like "evaluateDetectionPrecision” and “evaluateDetectionMissRate” to help you evaluate the performance of your object detector.
Clarifying the Sequence of using AlexNet with Faster R-CNN to monitor/detect your own image data
11 views (last 30 days)
I am currently starting a project with classifying images and object detection. This is my first time exploring the world of deep learning, so I just want to ensure that I got the sequence right when I'm coding it. For this project, I want to classify 5 different types of flowers and should be able to detect them in real-time with a camera. I would most likely have 1000 images per different types of flowers that are sized 277x277.
Below is the sequence that I came up with for my deep learning project:
First Part: Using AlexNet to classify my image data set.
- Load my training images.
- Split data into training and test set.
- Load Pre-trained Network (AlexNet)
- Modify Pre-trained Network to recognize only 5 image class
- Perform transfer learning
- Set a custom read function where it simply resizes the input images to 277x277
- Train the network
- Test the network performance to check the accuracy
Second Part: Using Faster R-CNN for Object Detection
- Create the ground truth table of my image dataset by using the Image Labeler App
- Train the Faster R-CNN with trainFasterRCNNObjectDetector command
- Use detect command to run the Fast R-CNN Object Detector
- Display the result with the object annotation command.
Below is just a visual example on what the output should look like:
Note: Please tell me if I'm missing any steps or if there is something wrong with the sequence. Also, I would appreciate it if you guys would give me more tips for a newbie like me.
Neha on 30 Aug 2023
Your sequence for the deep learning project looks generally correct. However, I would like to provide few additional tips and suggestions to help improve the performance of the model:
Hope this helps!