How can I improve the registration of my own data

[xl232lx]

Hellow author, I used demo.py to test my own src.npz and ref.npz files, and directly used geotransformer-3dmatch.pth.tar, but the effect was not very good, may I ask how I should improve the registration effect? In addition, this gt.npy file is used to calculate the error, will not affect the registration result?

[ChenxiWang09]

For gt.npy, as the author answered in other questions before, the demo.py is set to train, not predict, so you need to add model.eval() to the demo.py to make it run in predict mode, and then the gt will not affect the result.

[xl232lx]

Thank you for your reply!

[CossellCollege]

Hellow author, I used demo.py to test my own src.npz and ref.npz files, and directly used geotransformer-3dmatch.pth.tar, but the effect was not very good, may I ask how I should improve the registration effect? In addition, this gt.npy file is used to calculate the error, will not affect the registration result?

I have trouble converting data formats, could you tell me how is your data format conversion, I only have the location information of the text file into the npy file n × 3, but the error

[xl232lx]

Hellow author, I used demo.py to test my own src.npz and ref.npz files, and directly used geotransformer-3dmatch.pth.tar, but the effect was not very good, may I ask how I should improve the registration effect? In addition, this gt.npy file is used to calculate the error, will not affect the registration result?

I have trouble converting data formats, could you tell me how is your data format conversion, I only have the location information of the text file into the npy file n × 3, but the error

Hello, my file format is pcd, and then convert to npy file as input

[CossellCollege]

Hellow author, I used demo.py to test my own src.npz and ref.npz files, and directly used geotransformer-3dmatch.pth.tar, but the effect was not very good, may I ask how I should improve the registration effect? In addition, this gt.npy file is used to calculate the error, will not affect the registration result?

I have trouble converting data formats, could you tell me how is your data format conversion, I only have the location information of the text file into the npy file n × 3, but the error

Hello, my file format is pcd, and then convert to npy file as input

Did you adjust the scale of origin pointcloud into the voxel size in 3DMatch (2.5cm)?

[xl232lx]

Hellow author, I used demo.py to test my own src.npz and ref.npz files, and directly used geotransformer-3dmatch.pth.tar, but the effect was not very good, may I ask how I should improve the registration effect? In addition, this gt.npy file is used to calculate the error, will not affect the registration result?

I have trouble converting data formats, could you tell me how is your data format conversion, I only have the location information of the text file into the npy file n × 3, but the error

Hello, my file format is pcd, and then convert to npy file as input

Did you adjust the scale of origin pointcloud into the voxel size in 3DMatch (2.5cm)?

Hello, I did not change the point cloud file, my point cloud file resolution is 5cm.

[CossellCollege]

Hellow author, I used demo.py to test my own src.npz and ref.npz files, and directly used geotransformer-3dmatch.pth.tar, but the effect was not very good, may I ask how I should improve the registration effect? In addition, this gt.npy file is used to calculate the error, will not affect the registration result?

I have trouble converting data formats, could you tell me how is your data format conversion, I only have the location information of the text file into the npy file n × 3, but the error

Hello, my file format is pcd, and then convert to npy file as input

Did you adjust the scale of origin pointcloud into the voxel size in 3DMatch (2.5cm)?

Hello, I did not change the point cloud file, my point cloud file resolution is 5cm.

ok，thanks a lot!

[WishAllTheBest]

For gt.npy, as the author answered in other questions before, the demo.py is set to train, not predict, so you need to add model.eval() to the demo.py to make it run in predict mode, and then the gt will not affect the result.

Hello, how to add model.eval() to the demo.py I have no idea on this

[xl232lx]

For gt.npy, as the author answered in other questions before, the demo.py is set to train, not predict, so you need to add model.eval() to the demo.py to make it run in predict mode, and then the gt will not affect the result.

Hello, how to add model.eval() to the demo.py I have no idea on this

You can try it this way！