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
964 stars 64 forks source link

Will train with Depth Anything V2？ #38

Open libohit opened 4 months ago

gituser123456789000 commented 3 months ago

Second this... I couldn't get PatchFusion running locally, but I also request Depth Anything v2 support

zhyever commented 3 months ago

Hi guys, I think the PatchFusion is not the best solution for the hr depth estimation. We now have PatchRefiner with better performance. I will try to support Depth Anything v2 in PatchRefiner.

One more thing to be noticed: The PatchFusion uses the Swin Transformer block which is hard to be converted to onnx model. So it might be bad. In PatchRefiner, we drop it off and the model has even better performance.

gituser123456789000 commented 3 months ago

Hi guys, I think the PatchFusion is not the best solution for the hr depth estimation. We now have PatchRefiner with better performance. I will try to support Depth Anything v2 in PatchRefiner.

One more thing to be noticed: The PatchFusion uses the Swin Transformer block which is hard to be converted to onnx model. So it might be bad. In PatchRefiner, we drop it off and the model has even better performance.

Thank you for your responses. I'm interested to see results from PatchRefiner (and using Depth-Anything v2 along with whatever else may produce good results). I could never get PatchFusion running on Windows, so I hope the Windows install will be much easier with PatchRefiner. Thank you for your work.