search

haoyu94 / RoITr

Rotation-Invariant Transformer for Point Cloud Matching
MIT License
97 stars 13 forks source link

Can this work generalize to real-world scene? #13

Open mooncake199809 opened 3 months ago

mooncake199809 commented 3 months ago

Nice work. I am trying to use your algorithm to register two point clouds captured in real scenes. Can we directly utilize your work to register two point clouds in real scenes without retraining?

rjnps commented 3 months ago

Hi @mooncake199809, have you tried this with real world data, or any other dataset not used in the paper? How is its generalization capability?

mooncake199809 commented 3 months ago

Hi @mooncake199809, have you tried this with real world data, or any other dataset not used in the paper? How is its generalization capability?

hellow, it is OK

k794102577 commented 3 months ago

Can you briefly describe how you did that and what data set you used

mooncake199809 commented 2 months ago

Can you briefly describe how you did that and what data set you used

I use the model trained on the KITTI dataset and establish matches between two point clouds captured by ourselves.

k794102577 commented 2 months ago

How should kitti's config.yaml be set? When I am training kitti, loss nan will appear. Could you please provide the training hyperparameters in detail