search

LiyaoTang / ERDA

All Points Matter: Entropy-Regularized Distribution Alignment for Weakly-supervised 3D Segmentation (NeurIPS 2023)
MIT License
27 stars 2 forks source link

How to visualize the results? #4

Closed LinLin1031 closed 10 months ago

LinLin1031 commented 11 months ago

Thanks for your wonderful work. I would like to know how to visualize the results after validation or test? @LiyaoTang

LiyaoTang commented 11 months ago

Hi,

Thanks for your interest. But could you be more specific? For example, what would you want to visualize, and what type of visualization do you want to see?

Best, Liyao

LinLin1031 commented 11 months ago

Thank you for your prompt reply, before this issue, I encountered the following problem: I can train and output the accuracy results normally when the mode is train. But when mode is val, running 'python main.py' gives the following error: image The beginning part of the error output above, and the end part of the error output below: @LiyaoTang image

LiyaoTang commented 11 months ago

Hi,

As indicated by the error messages, you are trying to use uninitialized values, which indicates that the checkpoint that you specify to load is different from the model that you specify to use. Please double-check if you are loading the correct checkpoint.

Best, Liyao

LinLin1031 commented 11 months ago

Thanks for the advice, I've fixed the problem. Back on topic, I would like to ask for advice on how to visualize the results of a validation or test. Because currently I can only get quantitative results like the below figure but not qualitative results. @LiyaoTang

image

LiyaoTang commented 11 months ago

Hi,

Good to know that. Yes, we do not include the visualization code in the release, because it is coupled with several other projects and is hard to clean. However, we mostly adapt from the official RandLA-Net repo for visualization tools, so you could check out there if you want to do visualization.

Best, Liyao

LinLin1031 commented 11 months ago

I found some functions related to visualization in the utils/tester.py you provided. However I didn't find any place to call them. So can I just call them directly in main.py to visualize? @LiyaoTang

LinLin1031 commented 11 months ago

I found some functions related to visualization in the utils/tester.py you provided. However I didn't find any place to call them. So can I just call them directly in main.py to visualize? @LiyaoTang

I look forward to you answering my simple question at your convenience. Thank you very much! @LiyaoTang

LiyaoTang commented 11 months ago

Hi,

I do not think there is any visualization I wrote in the utils/tester.py, though there might be some leftover codes from KPConv / Closerlook3D.

But yes, you can certainly leverage those codes for your visualization purposes.

Best.