TonojiKiobya
TonojiKiobya
I only see script for training and demo but no script for testing the whole thing to get AP values of each category and the mAP for the whole thing...
Go to the trainval_net and look on the code where your backbone is (let say vgg16 or res101) and set pretrained to False. But I think this will take you...
Because we are using pretrained weights(resnet101_caffe.pth), so in essence we are doing transfer learning, hence it means that all layers have already been trained on imagenet and now we want...
Also check your anchor_scales in the RPN, if you have many it increases the complexity of your model, try to decrease the number of anchor_scales let say if you have...
I am also waiting for the information where the loss function was implemented!
Try changing the learning rate, decrease it or increase it by factor of 10 and see