CharlesShang / TFFRCNN

FastER RCNN built on tensorflow
MIT License
875 stars 418 forks source link

How to use earlier convolutional layers of VGG16 as input to the RPN? #85

Open hadign20 opened 7 years ago

hadign20 commented 7 years ago

I wanted to run faster rcnn on a custom dataset other than pascal VOC. But the VGG16 pretrained model isn't suitable for all sorts of data. So if I wanted to use another model or input the conv3_3 and conv4_3 activation maps into the RPN module, instead of conv5_3. In which file should I do this? I'm new to deep learning with tensorflow and would appreciate an expert's opinion on this. thanks