Label issue of training a faster R-CNN deep learning object detector
2 views (last 30 days)
Hi all! I met an issue of labeling the ROI region for training data. As different network requires different input size, once we resize the input image, the size of bouncing boxes will also change. I used the Image Labler to label the bouncing boxes for each traning image. The initial image size is 224*224 for Resnet50. If I want to use other networks such as AlexNet which requires 227*227 input size, does it mean that I have to relabel all the training images once again? Or is there any other method to adjust the size of bouncing boxes for the new input size?
Kritika Bansal on 2 Aug 2019
You can move the labels to the workspace and manipulate them according to your required size instead of labeling the whole data again.