pris-cv / pmg-v2

Code release for "Progressive Learning of Category-Consistent Multi-Granularity Features for Fine-Grained Visual Classification"

PMG-V2

Code release for "Progressive Learning of Category-Consistent Multi-Granularity Features for Fine-Grained Visual Classification" (IEEE T-PAMI 2021).

It is the extended version of "Fine-Grained Visual Classification via Progressive Multi-Granularity Training of Jigsaw Patches" (ECCV2020): https://github.com/PRIS-CV/PMG-Progressive-Multi-Granularity-Training

Requirement

python >= 3.6

PyTorch >= 1.3.1

torchvision >= 0.4.2

pandas >= 0.23.0

Training

1. Download datatsets for FGVC (e.g. CUB-200-2011, NA-Birds, Standford Cars, FGVC-Aircraft, etc) and organize the structure as follows:

data
├── bird_train.txt
├── bird_text.txt
├── train
│   ├── Black_Footed_Albatross_0065_796068.jpg
|   ├── Black_Footed_Albatross_0042_796071.jpg
|   ├── Black_Footed_Albatross_0090_796077.jpg
│   └── ...
└── test
    ├── Black_Footed_Albatross_0046_18.jpg
    ├── Black_Footed_Albatross_0002_55.jpg
    ├── Black_Footed_Albatross_0085_92.jpg
    └── ...

2. Train from scratch with train.py.

Contact

Thanks for your attention! If you have any suggestion or question, you can leave a message here or contact us directly:

• [email protected]
• [email protected]

pmg-v2's Issues

Question about the number of categories in nabirds

First of all, thank you for your novel work。
I have a problem when I train with the nabirds dataset as required by the dataset format, an exception occurs when setting the number of categories to 555. I would like to know if you are having this problem or if you have set a different number of categories。

great work, but trainling slowly，and can I only use the last branch stage for predict and abandon other branch?