Basically, I added a method to the occ_grid estimator to mark as empty the cells not covered by any mask (with a consensus_ratio), then apply a small dilation.
The implementation is similar to the already present mark_invisible_cells.
Even if the speed seems to not be concerned, this feature can help to avoid floaters in regions partially covered by the images.
If it seems a reasonable feature to you, please check it and tell me if it is necessary to change/improve the code.
Thanks you very much, and congratulations on the great work you made so far.
Hello @liruilong940607.
I added the feature described in VaxNeRF: Revisiting the Classic for Voxel-Accelerated Neural Radiance Field.
Basically, I added a method to the
occ_gridestimator to mark as empty the cells not covered by any mask (with aconsensus_ratio), then apply a small dilation. The implementation is similar to the already presentmark_invisible_cells.Even if the speed seems to not be concerned, this feature can help to avoid floaters in regions partially covered by the images.
If it seems a reasonable feature to you, please check it and tell me if it is necessary to change/improve the code. Thanks you very much, and congratulations on the great work you made so far.