Jianyuan Wang
Jianyuan Wang
In theory it should be okay
Hi, The tracker will predict for all queries, and then you can use the visibility score (and confidence score) to filter out those invisible ones
Hey I think it is definitely possible but did not try it before
Hi @li1u, thanks for the question! Just to clarify — in general, bundle adjustment (BA) jointly optimizes both (a) 3D point positions and (b) camera poses to minimize reprojection error....
Hi, if you do not use_ba, the results of demo_colmap should be exactly the same to demo_gradio.py. The only difference is, in demo_colmap we use a strict conf thres to...
Hi I guess this can solve your problem? Basically the padded areas bring some noisy points (but usually they should have very low confidence score right?) https://github.com/facebookresearch/vggt/pull/201
I see thanks for the feedback
It will be released with our next work
Yeah I think your understanding is correct. While for your example, could you try to adjust the confidence score a bit?
Did you get the number here? https://github.com/facebookresearch/vggt/blob/059c288227aa00faaf1a4ecf3f4f3be9b40aa467/training/data/datasets/vkitti.py#L110-L116