search

chzhang18 / StereoEchoes

Stereo Depth Estimation with Echoes at ECCV 2022
8 stars 1 forks source link

About Training strategies #3

Open cocowy1 opened 1 year ago

cocowy1 commented 1 year ago

Congratulations! I think your work opens a new path for stereo matching （multi-modal ), which can be applied for embodied agents. By the way, could you offer train.py code and the pretrained models about these two datasets? I want to reproduce your work, and why is Depth = 5 / Disparity ? The paper doesn't seem to mention the formula. To be honest, I think you can provide more training details so that more people follow your work more easily.In this way, the influence of this article in the community can be expanded.

chzhang18 commented 1 year ago

@cocowy1 Thanks for your attention! The training code may be released in the future. Unfortunately, there are still many things at hand.

The formula is the transformation between the depth and disparity. "To simulate camera baselines and focal lengths, we directly convert the depth Z to disparity D with D = Zmax/Z", where Zmax is set to 5. You can refer to the subsection "Stereo Images Synthesis" in Sec.4 of our paper for details.