search

zhyever / PatchFusion

[CVPR 2024] An End-to-End Tile-Based Framework for High-Resolution Monocular Metric Depth Estimation
https://zhyever.github.io/patchfusion/
MIT License
926 stars 62 forks source link

Quantitative comparison setting, and MVS-Synth dataset #39

Open Byeongjun1022 opened 3 weeks ago

Byeongjun1022 commented 3 weeks ago

  1. Can I know the setting for getting the same quantitative result with the paper?

  2. Do you have a plan to provide the patchfusion network trained with MVS-Synth? or training code for training with MVS-Synth? As far as I see, 'splits/u4k/train.txt' only contains UnrealStereo4K dataset.

Thank you for releasing the code