jacxue / crossmodalflow

CrossModalFlow

Pytorch implementation of Promoting Single-Modal Optical Flow Network for Diverse Cross-modal Flow Estimation (AAAI 2022)

Usage

Download the pre-trained model, and put it in the 'pre_trained' folder.

baidu yun access code: sztg

You can check 'run_model.py' for detailed usage.

Evaluation

Prepare the corresponding dataset and modify the path in 'evaluate_dataset.py'.

RGBNIR-Stereo

TriModalHuman

KITTI

Cross-KITTI access code: sztg

Then run evaluate_for_crossKitti.py / evaluate_for_rgbnir_stereo.py / evaluate_for_trimodalhuman.py.

Training

Prepare the dataset and modify the path in 'dataset.py'.

YoutubeVOS

Run 'train.py' to train.

中文说明

使用：

下载预训练模型，并放入pre_trained文件夹： 百度云 提取码 sztg

参考run_model.py的用法

评估：

准备相应的数据集，修改evaluate_dataset.py中的路径

RGBNIR-Stereo

TriModalHuman

KITTI

Cross-KITTI 提取码 sztg

运行evaluate_for_crossKitti.py / evaluate_for_rgbnir_stereo.py / evaluate_for_trimodalhuman.py

训练：

准备数据集，并修改dataset.py中的数据集路径

YoutubeVOS

运行train.py进行训练

Acknowledgments

Parts of the code are borrowed from: FlowNet2 PWC-Net RAFT