ECCV 2024 OOD-CV Workshop SSB Challenge (Open-Set Recognition Track) - 1st Place

This repository contains the code and resources for Intellindust-AI-Lab, which achieved the top performance in the ECCV 2024 OOD-CV Workshop SSB Challenge.

Table of Contents

1. Overview
2. Visual Results
3. SURE
4. Installation
5. Quick Start
6. Citation

Overview

An overview of our approach is illustrated below. You can read the detailed report here: SURE-OOD.pdf

Visual Results

The visual results for open-set recognition are presented below:

SURE

The SURE-OOD method extends the existing SURE framework by incorporating advanced techniques for open-set recognition, achieving state-of-the-art results in the OOD-CV challenge.

For more details on the SURE framework, please refer to the paper

Installation

Environment Setup

To set up the environment, install the dependencies listed in requirements.txt using the following command:

pip install -r requirements.txt

Note: For testing with a resolution of 480, you will need to modify the PatchEmbed and _pos_embed functions in the timm library. See Support.md for detailed instructions.

Dataset Preparation

Download the ImageNet-1k and ImageNet-21k validation sets using the Semantic Shift Benchmark (SSB) API. Make sure to specify the correct dataset path during training.

Quick Start

Training

To train the model, run the following script:

bash run/run_deit_Inet1k.sh

This will start the training process using the pre-configured parameters and the official pretrained model. The trained model will be saved to the specified directory.

Testing

After training, evaluate the model using the following script:

bash run/run_eval_TTA.sh

You can fuse the results and print them using:

python metric_result_GradNorm.py --result_dir path/to/

We provide a fine-tuned model: DeiT III-Base.

Citation

If our project is helpful for your research, please consider citing :

@InProceedings{Li_2024_CVPR,
    author    = {Li, Yuting and Chen, Yingyi and Yu, Xuanlong and Chen, Dexiong and Shen, Xi},
    title     = {SURE: SUrvey REcipes for building reliable and robust deep networks},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2024},
    pages     = {17500-17510}
}

and our challenge report:

@article{Li2024sureood,
    author    = {Li, Yang and Sha, Youyang and Wu, Shengliang and Li, Yuting and Yu, Xuanlong and Huang, Shihua and Cun, Xiaodong and Chen,Yingyi and Chen, Dexiong and Shen, Xi},
    title     = {SURE-OOD: Detecting OOD samples with SURE},
    month     = {September}
    year      = {2024},
}