search

PruneTruong / DenseMatching

Dense matching library based on PyTorch
GNU Lesser General Public License v2.1
689 stars 81 forks source link

Is the Warp Consistency Loss be suitable for small motion dataset? #15

Closed chenxingxin-star closed 2 years ago

chenxingxin-star commented 2 years ago

Hi, Thanks for this wonderful project! I want to train an optical flow network on my own dataset, but the view point changes in my dataset is small , Can the WarpC loss get a good result in such dataset? I found the dataset used in this project have a large view point changes.

PruneTruong commented 2 years ago

Hi, sure you can define yourself the strength of the transformation. The only limitation is, if the warp is all 0, then the W-bipath constraint simplifies to the forward-backward consistency constraint, which has a degenerate zero flow solution.

To sample transformations with small magnitudes, i would recommend sampling the parameters from a gaussian rather than uniformly. You can choose the magnitude of the parameters yourself.