ethnhe / FFB6D

[CVPR2021 Oral] FFB6D: A Full Flow Bidirectional Fusion Network for 6D Pose Estimation.
MIT License
299 stars 71 forks source link

test_occ not implemented? #8

Open CandiceD17 opened 3 years ago

CandiceD17 commented 3 years ago

While I run FFB6D on Linemod, I noticed that validation on occluded Linemod was not implemented in your code. I am wondering if I missed your implementation of the occluded Linemod somewhere, or will you add this feature for validation in the future?

ethnhe commented 3 years ago

That part of code hasn't been reconstructed for open source. I will add it to my to-do list as I'm busy with other projects these days.

Mr2er0 commented 3 years ago

I found that Linemod_preprocessed has processed LM_O information included mask_all and gt.yml in 02. Do you use this information for testing?

Mr2er0 commented 3 years ago

In addition, I would like to confirm whether the published pre-training weights can get the LM_O results in the paper. If no, can you publish the pre-training weights for the LM_O. Thank you a lot.

taeyeopl commented 3 years ago

As I understood, usually LM_O used the same weights as LM, because it doesn't have training data for LM_O. @ethnhe Does it right?? @Mr2er0 Have you evaluated the LM_O?? I also found --test_occ, the function doesn't work. https://github.com/ethnhe/FFB6D/blob/3579a69df27451f74f3abc06964bba1bc7d40605/ffb6d/train_lm.py#L84

Mr2er0 commented 3 years ago

As I understood, usually LM_O used the same weights as LM, because it doesn't have training data for LM_O. @ethnhe Does it right?? @Mr2er0 Have you evaluated the LM_O?? I also found --test_occ, the function doesn't work.

https://github.com/ethnhe/FFB6D/blob/3579a69df27451f74f3abc06964bba1bc7d40605/ffb6d/train_lm.py#L84

I write my own code for test LM_O. But i find that it works badly in the symmetrical objects while others work normally

William-Klay commented 2 years ago

As I understood, usually LM_O used the same weights as LM, because it doesn't have training data for LM_O. @ethnhe Does it right?? @Mr2er0 Have you evaluated the LM_O?? I also found --test_occ, the function doesn't work. https://github.com/ethnhe/FFB6D/blob/3579a69df27451f74f3abc06964bba1bc7d40605/ffb6d/train_lm.py#L84

I write my own code for test LM_O. But i find that it works badly in the symmetrical objects while others work normally

Hi, I am trying to test conduct testing for LM_O, but codes from PVNet do not work well, can you help me with that?

d-serafly commented 2 years ago

@Mr2er0 @William-Klay @taeyeop-lee Brothers, have you solved this problem? I also encountered the same problem when testing occ. In addition, do you have any ideas on invisible surface percentage?