Is there a way to use Stereocrafter with already processed DepthCrafter depth map videos or Depth Anything V2 videos?

[redman4585]

I figure it would be easier to have stereo generation with an already existing RGB video + Depth map video to free up VRAM

Like the first step has RGB video, plus the depthmap. and it just does the occlusion mask, and splatting right video.

[xiaoyu258]

Thanks for your suggestions.

It may require less VRAM by separating depth estimation and splatting. We will consider it in the next update.

[redman4585]

Thanks!

[xiaoyu258]

I have updated the "depth_splatting_inference.py" and it requires less VRAM now.

I created a new function "DepthSplatting" in "depth_splatting_inference.py" which you could use to splat the video with the video depth as input.

[redman4585]

Thanks!

Is this the proper input video type to use? A SBS video frame set up like this? Or do I keep it separate?

Are there any specific arguments to put in the inference code for it to work? Is it --video_depth and --depth_vis?

Or is the argument --video_depth the path to the depth video itself?