This is the faster rcnn based on Tensorflow and resnet50
- fork from the TFFRCN website is https://github.com/CharlesShang/TFFRCNN
- just change the python file factory.py ， __ init __.py ，networks.py in /lib/networks
- add resnet50 networks
- resnet101 hasn't been tested
You can find how to use it from the above website
the model file is convert from the caffemodel, you can download from the baiduyun
link: http://pan.baidu.com/s/1eSuUO1s pwd: 24cf
the fine-tune model link: http://pan.baidu.com/s/1nuUYfMh pwd：5bve
resnet101 imagenet file link http://pan.baidu.com/s/1i5odxNv pwd:d8y4
USing voc07_trainval to train the Resnet50 and test on the voc07_test
The result is :
Mean AP = 0.7124
AP for aeroplane = 0.7801
AP for bicycle = 0.7931
AP for bird = 0.6836
AP for boat = 0.5750
AP for bottle = 0.4892
AP for bus = 0.8322
AP for car = 0.8412
AP for cat = 0.8380
AP for chair = 0.5186
AP for cow = 0.7550
AP for diningtable = 0.6369
AP for dog = 0.7861
AP for horse = 0.7966
AP for motorbike = 0.7692
AP for person = 0.7765
AP for pottedplant = 0.4454
AP for sheep = 0.7131
AP for sofa = 0.7074
AP for train = 0.8150
AP for tvmonitor = 0.6953