search

sarafridov / K-Planes

Other
479 stars 45 forks source link

monocular real scene video #13

Open Spphire opened 1 year ago

Spphire commented 1 year ago

i tried to train from one monocular real scene video, i tried D-NeRF and DyNeRF config, but results are unsatisfying. has it the capacity to train from just one monocular real scene video? like nsff?

https://user-images.githubusercontent.com/56157591/227219365-2d6bb06c-b837-489a-b85f-792fcf040071.mp4

https://user-images.githubusercontent.com/56157591/227219759-cbae28c8-55d8-47ff-82b0-a6bc6f22f1c4.mp4

sarafridov commented 1 year ago

I'm not sure why but I'm unable to play the attached videos, so it's hard to debug the issue. In general K-Planes can train on monocular datasets, but the quality depends on the relative speed of the camera movement compared to the object movement; if the object moves too fast relative to the camera then we will produce blurry results due to lack of supervision. The following paper goes more in depth on monocular datasets, if it's useful for reference: https://hangg7.com/dycheck/