Comparison with ZoeDepth and Diffusion for Metric Depth (DMD).

[tcourat]

Hi,

It would be nice to compare your results with DMD ( Zero-Shot Metric Depth with a Field-of-View Conditioned Diffusion Model ) which also claim to be SOTA for zero shot metric depth.

I tried to compare directly the DMD (that follow ZoeDepth evaluation procedure) and your paper, but it seems that you followed a different evaluation procedure than ZoeDepth. Because I do not find the same results between the ZoeDepth in your paper and the original ZoeDepth paper. For instance, you report ZoeDepth having a AbsRel=0.520 on SUN RGB-D (Table 5) while ZoeDepth (original paper, Table 3) reports having AbsRel=0.119 with the architecture ZoeD-M12-N .

[LiheYoung]

Actually, we use the same evaluation protocol as ZoeDepth, except the image resolution (since our patch size is 14, whereas ZoeDepth's is 16). As for the SUN RGB-D results, we can not obtain the results close to ZoeDepth. Similar observations are also reported in ZoeDepth issues: https://github.com/isl-org/ZoeDepth/issues/19

[tcourat]

Thanks for answering quickly.

This is weird because the DMD paper seemed to obtain results coherent with those from ZoeDepth (they have AbsRel=0.091) on SunRGBD.