Chris Choy

It seems that it works perfectly fine if you use the provided download script.

It's on https://chrischoy.github.io/publication/dgr/#supplementary-materials On Tue, Aug 25, 2020 at 6:43 AM BiaoBiaoLi wrote: > Hi chrischoy: Thanks your work,I want know where is the supplement? > I want to know...

I see that the table is not very self-explanatory. FCGF by itself is not a registration algorithm. We used the RANSAC like the other experiments for FCGF and it is...

Point clouds used in KITTI are much larger than pointclouds used in 3DMatch. There's no reciprocity test in FCGF.

This is due to the API change. https://github.com/chrischoy/FCGF/issues/38 From MinkowskiEngine v0.5, the quantization function returns coordinates and indices. Replace `sel0= ME.utils.sparse_quantize(xyz0 / voxel_size, return_index=True)` with `_, sel0= ME.utils.sparse_quantize(xyz0 / 0.05,...

It's used in https://github.com/chrischoy/FCGF/blob/master/lib/trainer.py#L329. Both transformation estimation and ContrastiveLossTrainer are kind of legacy code we do not use for training FCGFs.

The method used in `solve_linear_system` is a simpler version of FGR. For the final registration recall, we use the original authors matlab code that uses RANSAC as I have explained...

Yes, your first model was the one that I tried first and experimentally, I found that the current model works better. The diagram might not match the implementation as we...

The commercial depth sensors generate noisy samples that are not suitable for registration. A typical pipleline first integrate 50-100 frames into one point cloud. We also followed this procedure and...

For V100 speed being slower than 1080ti, use export OMP_NUM_THREADS=20 or lower.