inquiry on the future

[donrikk]

hi @LiheYoung love the work you are doing here! i was just reaching out to ask you for some insight on your possible future plans with depth anything. i know you'll be releasing giant model later on once you can do so, but i wanted to ask about the prospect of a V3 model at some point and if this is something you are actively working on what could we expect to be different going forward. as of right now v2 has very little problems in its depth estimation, the only thing it seems to have a problem with is blurred objects in scenes of a video, say a person in the foreground is blurred or an object in the midground is blurred the map will either set that object to the midground or worse to the back as a black in the map. ive been able to fix this by blending your v1 maps that dont have that problem with the v2 maps that do have that issue fixing most of the issue with some problem scenes still present but overall ignorable. if possible i think a blending of the v1 and v2s map logic would be an interesting project, since you already have 2 active models that can fix each others problems in my head it would make sense to then combine both of there logic together in a cohesive way. but im far from as knowledgeable as yourself so i dont know if that is possible for you to achieve. thank you in advance for any info you can give me.

[vfan26]

hi @LiheYoung love the work you are doing here! i was just reaching out to ask you for some insight on your possible future plans with depth anything. i know you'll be releasing giant model later on once you can do so, but i wanted to ask about the prospect of a V3 model at some point and if this is something you are actively working on what could we expect to be different going forward. as of right now v2 has very little problems in its depth estimation, the only thing it seems to have a problem with is blurred objects in scenes of a video, say a person in the foreground is blurred or an object in the midground is blurred the map will either set that object to the midground or worse to the back as a black in the map. ive been able to fix this by blending your v1 maps that dont have that problem with the v2 maps that do have that issue fixing most of the issue with some problem scenes still present but overall ignorable. if possible i think a blending of the v1 and v2s map logic would be an interesting project, since you already have 2 active models that can fix each others problems in my head it would make sense to then combine both of there logic together in a cohesive way. but im far from as knowledgeable as yourself so i dont know if that is possible for you to achieve. thank you in advance for any info you can give me.

Hello, I have also been troubled by a similar problem recently. The blurred edge of the foreground is always assigned the depth of the background. How do you mix the depth maps of V1 and V2 to solve this problem? Thank you for your specific suggestions.