Training faster r-cnn for INRIA dataset

[nimamahmoudi]

Hi,

I trained a model based on this great tutorial from deboc: https://github.com/deboc/py-faster-rcnn/blob/master/help/Readme.md and the training shows to be successful (loss is about 0.00XXX).

This is my training command: ./tools/train_faster_rcnn_alt_opt.py --gpu 0 --net_name INRIA_Person --weights data/imagenet_models/VGG_CNN_M_1024.v2.caffemodel --imdb inria_train --cfg experiments/cfgs/faster_rcnn_alt_opt.yml

but the problem is when I run the following, the mAP that I get is very low (~0.01): ./tools/test_net.py --gpu 0 --def ~/Documents/py-faster-rcnn/models/pascal_voc/INRIA_Person/faster_rcnn_alt_opt/faster_rcnn_test.pt --net ~/Documents/py-faster-rcnn/output/faster_rcnn_alt_opt/train/vgg_cnn_m_1024_rpn_stage1_iter_10000.caffemodel --imdb inria_train --cfg experiments/cfgs/faster_rcnn_alt_opt.yml

the structure of my files looks like this: ├── caffe-fast-rcnn │ ├── build -> .build_release │ ├── caffe.cloc │ ├── cmake │ ├── CMakeLists.txt │ ├── CONTRIBUTING.md │ ├── CONTRIBUTORS.md │ ├── data │ ├── distribute │ ├── docs │ ├── examples │ ├── include │ ├── INSTALL.md │ ├── LICENSE │ ├── Makefile │ ├── Makefile.config │ ├── Makefile.config.example │ ├── matlab │ ├── models │ ├── python │ ├── README.md │ ├── scripts │ ├── src │ └── tools ├── config.yml ├── data │ ├── cache │ ├── demo │ ├── faster_rcnn_models │ ├── imagenet_models │ ├── INRIA_Person_devkit │ ├── pylintrc │ ├── README.md │ └── scripts ├── experiments │ ├── cfgs │ ├── logs │ ├── README.md │ └── scripts ├── help │ ├── faster_rcnn_alt_opt.yml │ └── Readme.md ├── lib │ ├── datasets │ ├── fast_rcnn │ ├── Makefile │ ├── nms │ ├── pycocotools │ ├── roi_data_layer │ ├── rpn │ ├── setup.py │ ├── transform │ └── utils ├── LICENSE ├── models │ ├── coco │ ├── INRIA_Person │ ├── pascal_voc │ └── README.md ├── output │ └── default ├── README.md └── tools ├── compress_net.py ├── demo.py ├── eval_recall.py ├── _init_paths.py ├── _init_paths.pyc ├── README.md ├── reval.py ├── rpn_generate.py ├── test_net.py ├── train_faster_rcnn_alt_opt.py ├── train_net.py └── train_svms.py

Thanks for you help.

[nimamahmoudi]

anybody?

[deboc]

Hi, You are testing the rpn stage1 caffemodel network, right ? I think you want to test the vgg_cnn_m_1024_faster_rcnn_final.caffemodel instead

[sujeewa1]

When I am training the vgg_cnn_m_1024 models its gives GPU memory issue when loading the stage two.. I reduced the nuumber of iteration to 100 and still the error is there. My GPU is GTX1080. Can you please tell me what is this error as I think GTX1080 is enough to train the network..