Awesome

TPCV

code for CVPR23 Paper Learning optical expansion from scale matching

Requirements

The code has been tested with PyTorch 1.11.0 and Cuda 11.3.

conda create -n tpcv python=3.9
conda activate tpcv
pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 torchaudio==0.11.0 --extra-index-url https://download.pytorch.org/whl/cu113

Compile CUDA code for inverse optical flow

cd inverse_cuda
python setup.py install

Also need to install via pip

pip install matplotlib
pip install opencv-python
...

Dataset Configuration

To evaluate/train TPCV, you will need to download the required datasets.

• FlyingChairs
• FlyingThings3D
• Sintel
• KITTI

We recommend manually specifying the path in dataset_exp_orin.py, like def __init__(self, aug_params=None, split='kitti_test', root='/new_data/datasets/KITTI/training',get_depth=0): , because the automatic one often makes mistakes

You can create symbolic links to wherever the datasets were downloaded in the datasets folder

├── datasets
    ├── Sintel
        ├── test
        ├── training
    ├── KITTI
        ├── testing
        ├── training
        ├── devkit
    ├── FlyingChairs_release
        ├── data
    ├── FlyingThings3D
        ├── frames_cleanpass
        ├── frames_finalpass
        ├── optical_flow

Pretrained weights on KITTI

https://drive.google.com/drive/folders/1Ddh1HYKVo5CVITmLfV9_8uJKMI3HdePd?usp=share_link

Train on KITTI

python train.py --name raft-kitti_3D --stage kitti --validation kitti --restore_ckpt ../TPCV/checkpotins/kitti_3D_flow.pth --gpus 0 --num_steps 60000 --batch_size 2 --lr 0.000125 --image_size 320 960 --wdecay 0.0001 --gamma=0.85

Test on KITTI

python dc_flow_eval.py --model=../TPCV/checkpotins/kitti_3D_flow.pth --mixed_precision --start=0