search

humanpose1 / MS-SVConv

Compute descriptors for 3D point cloud registration using a multi scale sparse voxel architecture
MIT License
63 stars 7 forks source link

Can you give the data from 0.00-0.20 results? #10

Closed QWTforGithub closed 3 years ago

QWTforGithub commented 3 years ago

Hi,I would like to make some experimental comparisons according to your work. Could you give me the Feat Macthing Ratio from 0.00 to 0.20? The results I run by myself may not meet the requirements of the results you run, so I hope you can give me, thank you.

Looking forward your reply.

humanpose1 commented 3 years ago

Hi, Sure, but I'm not sure what you exactly want. You want the FMR of 3DMatch for MS-SVConv 3 scales (or ETH) ? Can you specify the dataset ? and the model ?

QWTforGithub commented 3 years ago

Hi, Sure, but I'm not sure what you exactly want. You want the FMR of 3DMatch for MS-SVConv 3 scales (or ETH) ? Can you specify the dataset ? and the model ?

Yes. May I venture to ask, has your article been accepted? If accepted, I need your data on 3DMatch(3 head). Of course, if you can provide 1 head of 3DMatch data that would be much better, I would appreciate it.

QWTforGithub commented 3 years ago

Hi, Sure, but I'm not sure what you exactly want. You want the FMR of 3DMatch for MS-SVConv 3 scales (or ETH) ? Can you specify the dataset ? and the model ?

Yes. May I venture to ask, has your article been accepted? If accepted, I need your data on 3DMatch(3 head). Of course, if you can provide 1 head of 3DMatch data that would be much better, I would appreciate it. By the way, do you know how to get the gradient of MinkowskiEngin in a certain layer? I have been troubled by this problem for a long time. The network built by MinkowskiEngin computs that loss propagates backward without calling the hook function bound to a certain layer, so I don't know how to get the corresponding gradient, any suggestions would be appreciated，thank you.

humanpose1 commented 3 years ago

For the second question, I'm not sure but have you tried model.__dict__ ? there is the attribute '_backward_hooks'. As for each minkowski convolution, there are an attribute kernel and bias which are torch Tensor. so it has a grad. I hope it can help you.

humanpose1 commented 3 years ago

Here, you have the results of MS-SVConv 3 scales on 3DMatch (here for MS-SVConv 1 scale ). the feature match ratio for each scene individually and the hit ratio also (after a symmetric test). You have other information as well (error of translation or error of rotation). NB: you can open it with pandas

df = pd.read_csv(your_csv_file)
print(df.groupby("name_scene").mean())

Does it help you ?

QWTforGithub commented 3 years ago

Here, you have the results of MS-SVConv 3 scales on 3DMatch (here for MS-SVConv 1 scale ). the feature match ratio for each scene individually and the hit ratio also (after a symmetric test). You have other information as well (error of translation or error of rotation). NB: you can open it with pandas

df = pd.read_csv(your_csv_file)
print(df.groupby("name_scene").mean())

Does it help you ?

Thank you for your thoughtful reply!