lvyilin / pytorch-fgvc-dataset

PyTorch custom dataset APIs -- CUB-200-2011, Stanford Dogs, Stanford Cars, FGVC Aircraft, NABirds, Tiny ImageNet, iNaturalist2017

PyTorch FGVC Dataset

This repo contains some unofficial PyTorch dataset APIs(mainly for Fine-Grained Visual Categorization task), which support automatically download (except large-scale datasets), extract the archives, and prepare the data.

Supported Datasets

• CUB-200-2011
• Stanford Dogs
• Stanford Cars
• FGVC Aircraft
• NABirds
• Tiny ImageNet
• iNaturalist 2017
• Oxford 102 Flowers
• Oxford-IIIT Pets
• Food-101

Usage

The code was tested on

• pytorch==1.4.0
• torchvision==0.4.1

Use them the similar way you use torchvision.datasets.

train_dataset = Cub2011('./cub2011', train=True, download=False)
test_dataset = Cub2011('./cub2011', train=False, download=False)

Contributing

Feel free to open an issue or PR.

License

MIT

pytorch-fgvc-dataset's Issues

[Breaking Change] CUB-200-2011 is now hosted on Google Drive

CUB-200-2011 is now hosted on Google Drive.

The official link is the same, but it redirects to this.

A simple solution to this is to use torchvision.utils.download_file_from_google_drive instead of torchvision.utils.download_url and set the correct file_id (1hbzc_P1FuxMkcabkgn9ZKinBwW683j45).