Xiaojie0903/genview_pretrained_models

GenView Pretrained Models

Model Name

GenView: Enhancing View Quality with Pretrained Generative (ECCV 2024)

Summary

This repository hosts pretrained models developed as part of the GenView framework, introduced in the ECCV 2024 paper GenView: Enhancing View Quality with Pretrained Generative Model for Self-Supervised Learning. These models are designed for visual representation tasks, including image classification, multimodal learning, and feature extraction. GenView leverages generative models to enhance self-supervised learning by improving view quality and diversity.

---

Table of Contents

1. Model Details
2. Evaluation
3. Citation
4. How to Download the Model

---

Model Details

Model Description

The GenView pretrained models include both convolutional architectures (e.g., ResNet50) and transformer-based architectures (e.g., ViT-B). These models utilize advanced self-supervised learning methods such as SimSiam, MoCo, and BYOL. By incorporating generative models for adaptive view generation, the framework delivers superior feature representations.

• Developed by: Xiaojie Li, Yibo Yang, Xiangtai Li, Jianlong Wu, Yue Yu, Bernard Ghanem, Min Zhang
• Funded by: Harbin Institute of Technology, Shenzhen; Peng Cheng Laboratory; KAUST; NTU
• Shared by: Xiaojie Li
• Model type: Self-supervised learning for vision tasks
• Language: Vision-focused (not language-specific)
• License: Apache 2.0

Model Sources

• Hugging Face Repository: GenView Pretrained Models
• GitHub Repository: GenView Official Code
• Paper: GenView: Enhancing View Quality with Pretrained Generative Model for Self-Supervised Learning (ECCV 2024)

---

Evaluation

Testing Data

Linear Probe evaluation was conducted using the ImageNet-1K dataset.

Metrics

The models were evaluated based on Top-1 accuracy.

Results

Method	Backbone	Pretraining Epochs	Linear Probe Accuracy (%)
MoCo v2 + GenView	ResNet-50	200	70.0
SwAV + GenView	ResNet-50	200	71.7
SimSiam + GenView	ResNet-50	200	72.2
BYOL + GenView	ResNet-50	200	73.2
MoCo v3 + GenView	ResNet-50	100	72.7
MoCo v3 + GenView	ResNet-50	300	74.8
MoCo v3 + GenView	ViT-S	300	74.5
MoCo v3 + GenView	ViT-B	300	77.8

---

Citation

If you use these models, please cite the GenView paper:

@inproceedings{li2023genview,
  author={Li, Xiaojie and Yang, Yibo and Li, Xiangtai and Wu, Jianlong and Yu, Yue and Ghanem, Bernard and Zhang, Min},
  title={GenView: Enhancing View Quality with Pretrained Generative Model for Self-Supervised Learning}, 
  year={2024},
  booktitle={Proceedings of the European Conference on Computer Vision},
  pages={306--325},
  publisher="Springer"
}

---

How to Download the Model

Downloading Models

To download models, use the following commands:

Option 1: wget

# Replace {MODEL_FILE} with the specific model file name
wget https://huggingface.co./Xiaojie0903/genview_pretrained_models/resolve/main/{MODEL_FILE}

Example:

wget https://huggingface.co./Xiaojie0903/genview_pretrained_models/resolve/main/mocov3_resnet50_8xb512-amp-coslr-100e_in1k_genview.pth

Option 2: Hugging Face Python API

from huggingface_hub import hf_hub_download

# Replace with your desired model file
file_path = hf_hub_download(
    repo_id="Xiaojie0903/genview_pretrained_models",
    filename="mocov3_resnet50_8xb512-amp-coslr-100e_in1k_genview.pth"
)
print(f"Model downloaded to {file_path}")