A pytorch implementation of Bilinear CNNs for Fine-grained Visual Recognition(BCNN).
- python 2.7
- pytorch 0.4.1
Step 1.
-
Download the vgg16 pre-training parameters. vgg16-download.
-
Download the CUB-200-2011 dataset. CUB-download
Step 2.
- Set the path to the dataset and vgg parameters in the code.
Step 3. Train the fc layer only. It gives 77.30% test set accuracy.
- python train_last.py
Step 4. Fine-tune all layers. It gives 84.40% test set accuracy.
- python train_finetune.py