What an awesome video! Great presentation, thank you for saving me the time to do this on my own... and money. This is the first video I've seen and I've already decided to subscribe!
You are right, depth anything v2 works great for images, though Apple's depth pro gives great results on some images, on some images it just fails completely making it not reliable.
I'm really enjoying your videos, Kevin! I was wondering if you've had a chance to explore upscaling algorithms like ESRGAN. A deep dive into this topic would be fantastic for the community.
Thanks for the video. Is it possible to convert the depth map prediction into a point cloud or any type of 3D reconstruction? I have tried using random and accurate camera intrinsics but the 3D data is distorted
Great video. I am working with a side view video of athletes running. I have pose keypoints and want to overlay a depth model to estimate keypoint depths. Which model works best?
I want to use depth-pro to directly obtain the depth value and then convert it into three-dimensional information to complete the distance measurement. Is this feasible?
Do you think that Depth Anything V2 will develop future models to handle the video flickering better? I'm guessing yes. It's a great model, particularly for single images, but as you pointed out it can produce flickering for videos as I've tested this out myself as well.
Thanks, Kevin. Another great video 🙂
Thank you! 😊
What an awesome video! Great presentation, thank you for saving me the time to do this on my own... and money. This is the first video I've seen and I've already decided to subscribe!
Thanks for the sub!
Very interested in this type of video. Hopefully you can continue with it.
Thanks! Will do!
excelente video!!!
Thanks!
You are right, depth anything v2 works great for images, though Apple's depth pro gives great results on some images, on some images it just fails completely making it not reliable.
Yeah good thing to test them out for your dataset!
I'm really enjoying your videos, Kevin! I was wondering if you've had a chance to explore upscaling algorithms like ESRGAN. A deep dive into this topic would be fantastic for the community.
Thanks! I’ll look into it
Thanks for the video. Is it possible to convert the depth map prediction into a point cloud or any type of 3D reconstruction? I have tried using random and accurate camera intrinsics but the 3D data is distorted
Yes it’s possible. If it’s distorted most likely your depth map is not accurate enough
Great video. I am working with a side view video of athletes running. I have pose keypoints and want to overlay a depth model to estimate keypoint depths. Which model works best?
I would try out Sapiens from meta
I want to use depth-pro to directly obtain the depth value and then convert it into three-dimensional information to complete the distance measurement. Is this feasible?
Is the depth value obtained in meters?
Yes meters. Should be possible
@@kevinwoodrobotics I directly return the depth value. After I converted the 3D data, it seems that the depth unit is feet. Maybe I made a mistake.
Do you think that Depth Anything V2 will develop future models to handle the video flickering better? I'm guessing yes. It's a great model, particularly for single images, but as you pointed out it can produce flickering for videos as I've tested this out myself as well.
I’ve seen some attempts at that but nothing officially from them. Hopefully they can combine with Depthcrafter
I tried depth anything v2 and metric3d for metric depth, and they both gave very inconsistent results
Yeah may need some fine tuning
what about stereo , make reviews
Will add that to my list, thanks!
You're showing the vae checkpoint for marigold and not the unet
Thanks for pointing that out