Pytorch Implementation for "Deep Patch Learning for Weakly Supervised Object Classification and Discovery" paper
PASCAL VOC2012 Testset
mAP: 0.90240
Class | aeroplane | bicycle | bird | boat | bottle | bus | car | cat | chair | cow |
---|---|---|---|---|---|---|---|---|---|---|
AP | 0.98400 | 0.92760 | 0.95630 | 0.93490 | 0.77990 | 0.92200 | 0.90910 | 0.97970 | 0.81800 | 0.90490 |
Class | diningtable | dog | horse | motorbike | person | pottedplant | sheep | sofa | train | tvmonitor |
AP | 0.79660 | 0.97180 | 0.96420 | 0.94030 | 0.97750 | 0.70770 | 0.92720 | 0.77180 | 0.97240 | 0.90220 |
- Compling libs for this framework
cd lib/model/
cd roi_align/ && ./make.sh
cd roi_pooling && ./make.sh
cd spmmax_pooling && ./make.sh
- train
python train.py --imageset [train, trainval] --basemodel [vgg, resnet34, resnet50] --data_dir <Data Directory Path>
- proposal
- densebox sampling
- selective search
written as a PyTorch Extension and supported CUDA
see: ./lib/model/spmmax_pooling
This project is under MIT Licence