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！

[mqxb9527]

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?

Hello, this is my first time to implement this work. How should I organize the weights structure of the file geotransformer-3dmatch.pth.tar when training 3dMatch? Thank you very much.

[jun3333333]

Have you prepared your dataset?I'd like to consult you about knowledge related to datasets. My email address is 2373178085@qq.com.

[guoyanzhongwgfufg]

I think ,run the demo for own datset. first thing match difference in owndate and 3dmatch dataset. My only 400-3000point face pointcloud far than room various scene(3dmatch scene） , at all not fit ，so must train you model and choose fit selfdata models. or else just a caorse registration and speed very slow.