LIFE: Lighting Invariant Flow Estimation

Project Page | Paper

LIFE: Lighting Invariant Flow Estimation
Zhaoyang Huang^*, Xiaokun Pan^*, Runsen Xu, Yan Xu, Ka Chun Cheung, Guofeng Zhang, Hongsheng Li

Requirements

conda env create -f environment.yaml
conda activate life

Demos

1. For single image warping, please refer to the notebooks/flow_warp.ipynb.
2. LIFE can robustly align a picture to an image under significant lighting variation. To synthesize an image with realistic lighting conditions, please refer to our previous work NIID-Net.
3. To play a movie on the paper, please refer to DEMO.md

Training

Comming soon.

Evaluation

KITTI

cd evaluation
python kitti_eval.py --model PATH_To_MODEL --version 2012/2015 --data_dir PATH_TO_KITTI_DATASET

Acknowledgements

Our neural network architecture is mainly built upon RAFT.