Open dgiunchi opened 2 months ago
As far as I know, the most messy case is that the motion part of the scene is too big (please refer to our supp). In my code, the time variable is mainly based on the hypothesis that you only take the monocular video as input. So the first frame of the video is assigned to timestamp 0, and the last frame of the video is assigned to timestamp 1.
As far as I know, the most messy case is that the motion part of the scene is too big (please refer to our supp). In my code, the time variable is mainly based on the hypothesis that you only take the monocular video as input. So the first frame of the video is assigned to timestamp 0, and the last frame of the video is assigned to timestamp 1.
I assume for stereoscopic capturing this would fail, correct?
I'm trying to generate the right inputs from colmap using nerfstudio starting from an mp4. At the end of the training when I visualize, it is a total mess. I noticed, making a comparison between the transforms.json from the dnerf and the custom one generated by colmap, that is missing the "time" variable. Did somebody have the same problem with custom data?