Self-produced Guidance for Weakly-supervised Object Localization
PyTorch implementation of "Self-produced Guidance for Weakly-supervised Object Localization", paper
Prerequisites
- Python 3.6+
- Pytorch ( >= 1.1)
- Python bindings for OpenCV
- tqdm
- tensorboardX
Data Preparation
Execution
- Train
git clone https://github.com/halbielee/SPG_pytorch.git
cd SPG_pytorch
# Before executing, please set the appropriate dataset path
bash script/train_[dataset].sh
- Evaluate
# Before executing, please set the appropriate dataset path
bash script/evaluate_[dataset].sh
Performance
We evaluate the trained model of SPG which the author provides in our code.
| Data | Top1 Cls(%) | Top5 Cls(%) | Top1 Loc(%) | Top5 Loc(%) | Gt-Known |
|---|---|---|---|---|---|
| ImageNet | 66.158 | 87.482 | 45.656 | 58.212 | 62.800 |
| CUB-200-2011 | 68.226 | 89.990 | 46.307 | 60.200 | 64.791 |
Qualitative Image
