https://github.com/SSDUT-Caiyq/UFG-NCD

(CVPR2024 Highlight) Novel Class Discovery for Ultra-Fine-Grained Visual Categorization (UFG-NCD)
https://github.com/SSDUT-Caiyq/UFG-NCD

gcd ncd ultra-fine-grained

Last synced: about 2 months ago
JSON representation

(CVPR2024 Highlight) Novel Class Discovery for Ultra-Fine-Grained Visual Categorization (UFG-NCD)

• Host: GitHub
• URL: https://github.com/SSDUT-Caiyq/UFG-NCD
• Owner: SSDUT-Caiyq
• Created: 2024-03-24T12:42:22.000Z (10 months ago)
• Default Branch: main
• Last Pushed: 2024-07-01T07:13:42.000Z (6 months ago)
• Last Synced: 2024-08-02T15:29:57.795Z (5 months ago)
• Topics: gcd, ncd, ultra-fine-grained
• Language: Python
• Homepage:
• Size: 720 KB
• Stars: 12
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# UFG-NCD

Official pytorch implementation of our paper: [*Novel Class Discovery for Ultra-Fine-Grained Visual Categorization*](https://openaccess.thecvf.com/content/CVPR2024/papers/Liu_Novel_Class_Discovery_for_Ultra-Fine-Grained_Visual_Categorization_CVPR_2024_paper.pdf) (CVPR2024 **Highlight**)

![overview](./assets/overview.jpg)

## Environment :snake:

Our implementation is based on [uno](https://github.com/DonkeyShot21/UNO), while logging is performed using `wandb`, we use `conda` to create the environment and install the dependencies.

Follow the commands below to setup environment.

```bash

# create environment

conda create -n rapl python=3.8

conda activate rapl

# choose the cudatoolkit version on your own

conda install pytorch==1.7.1 torchvision==0.8.2 cudatoolkit=11.0 -c pytorch

# install other dependencies

pip install tqdm wandb scikit-learn pandas pytorch-lightning==1.1.3 lightning-bolts==0.3.0 

# create checkpoints directory

mkdir checkpoints

```

## Datasets :floppy_disk:

We use Ultra-FGVC as our main experiment datasets, specifically the `SoyAgeing-{R1, R3, R4, R5, R6}`.

You can download the datasets [here](https://github.com/XiaohanYu-GU/Ultra-FGVC), also make sure you have changed the datasets root that defined in [`config.py`](./config.py).

## Supervised Learning :star2:

```bash

python supervised_learning.py --dataset SoyAgeing-R1 --task ncd

```

## Discovery Learning :sparkles:

```bash

python discovery_learning.py --dataset SoyAgeing-R1 --task ncd --pretrained ncd-supervised-SoyAgeing-R1-pc2.0-cra0.6-reg1.0.pth

```

## Citation :clipboard:

If you find our work helpful, please consider citing our paper:

```tex

@InProceedings{Liu_2024_CVPR,

    author    = {Liu, Yu and Cai, Yaqi and Jia, Qi and Qiu, Binglin and Wang, Weimin and Pu, Nan},

    title     = {Novel Class Discovery for Ultra-Fine-Grained Visual Categorization},

    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},

    month     = {June},

    year      = {2024},

    pages     = {17679-17688}

}

```

## Acknowledgements :gift:

- [UNO](https://github.com/DonkeyShot21/UNO)

- [Ultra-FGVC](https://github.com/XiaohanYu-GU/Ultra-FGVC)