search

rbgirshick / py-faster-rcnn

Faster R-CNN (Python implementation) -- see https://github.com/ShaoqingRen/faster_rcnn for the official MATLAB version
Other
8.11k stars 4.11k forks source link

Testing Pretrained Imagenet Models #242

Open purple-panda opened 8 years ago

purple-panda commented 8 years ago

Hello!

This is my first time experimenting with caffe so I apologize in advance if this is very simple.

I was able to successfully run the demo. I was also able to successfully run ./tools/test_net.py --def models/pascal_voc/VGG16/faster_rcnn_alt_opt/faster_rcnn_test.pt --net data/faster_rcnn_models/VGG16_faster_rcnn_final.caffemodel --vis

Now I'm trying to test the pre-trained imagenet models provided. I've tried the following commands ./tools/test_net.py --def models/pascal_voc/VGG16/faster_rcnn_alt_opt/faster_rcnn_test.pt --net data/imagenet_models/VGG16.v2.caffemodel --vis

./tools/test_net.py --def models/pascal_voc/VGG16/faster_rcnn_end2end/test.prototxt --net data/imagenet_models/VGG16.v2.caffemodel

The script runs without error but it looks like the VGG model isn't actually detecting anything. Should I be using a different prototext file? What would that be?

IdiosyncraticDragon commented 8 years ago

The VGG16.v2.caffemodel trained for classification, it surely can not detect any things.

ednarb29 commented 8 years ago

Right, the model is just for initialization, that you don't need to train from scratch. There are some papers out there that describe learning on pre-trained models due to their generalisation in the early layers.

ashutalewar commented 7 years ago

@purple-panda - were you able to run the demo using imagenet pretrained VGG model? I am also trying the same thing and facing a similar issue.