README.md

Efficient Event Camera Data Pretraining with Adaptive Prompt Fusion
_{PyTorch Implementation of the ICCV 2025 paper}
Supp

Getting Started

Requirement

• python 3.8
• numpy 1.24.1
• torch 1.11.0
• torchvision 0.10.1
• pytorch_lightning 1.6.4
• einops 0.4.0
• timm 0.9.2
• flatten_dict 0.4.2

Datasets

• N-ImageNet dataset. The expected file sturctures are shown below.

└───N_Imagenet
    └───/extracted_train/n******/***.npz
    │
    └───/extracted_val/n******/***.npz

Pre-training

• Change the config accordingly
• Run the following code:

python train.py --opt config/ours/pr_vits.yml --gpus 8 --num_nodes 1

Acknowledgement

Most of the code is borrowed from:

• https://github.com/82magnolia/n_imagenet
• https://github.com/Yan98/Event-Camera-Data-Pre-training
• https://github.com/Yan98/Event-Camera-Data-Dense-Pre-training

Citation

If you find this code useful, please consider citing:

@inproceedings{liang2025efficient,
  title={Efficient Event Camera Data Pretraining with Adaptive Prompt Fusion},
  author={Liang, Quanmin and Li, Qiang and Liu, Shuai and Cao, Xinzi and Lu, Jinyi and Yang, Feidiao and Zhang, Wei and Huang, Kai and Tian, Yonghong},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={8656--8667},
  year={2025}
}