search

chensong1995 / HybridPose

HybridPose: 6D Object Pose Estimation under Hybrid Representation (CVPR 2020)
MIT License
412 stars 64 forks source link

About eggbox in LM-O #75

Open weidu3 opened 1 year ago

weidu3 commented 1 year ago

Hi, I found the eggbox in the LM-O dataset has been suppressed by a red car, just like the picture below. . So I'm curious, eggbox tested on the LM-O dataset, did it use the original data of LM-O? 0012

chensong1995 commented 1 year ago

Hi weidu3,

Thanks for your interest in our work! As the name suggests, the Linemod Occlusion dataset involves many examples with medium to severe occlusions. In our experiments, we seek to predict object poses from the original RGB images provided by the dataset

I hope this helps! Let me know if you have further concerns.

weidu3 commented 1 year ago

Hi chensong, Thank you especially for your reply. My eggbox's test effect on the LM-O dataset of pvn3d and ffb6d is far from the adds indicator in the paper, so I was wondering if the eggbox model is different between LM and LM-O. So I wonder if you understand the situation!

weidu3 commented 1 year ago

So, I wonder if symmetrical objects like Glue and EggBox require additional processing when evaluating on the LM-O dataset?

chensong1995 commented 1 year ago

Hi weidu3,

Symmetric objects like glue and eggbox are evaluated using the ADD-S metric instead of the ADD metric (relevant code in our repository). This is also the convention among most 6D pose estimation experiments. I hope this clarifies! Let me know if you have further concerns.

weidu3 commented 1 year ago

I confirm that the evaluation metrics are fine, and I want to confirm that the LMO dataset used in many papers is the LMO in BOB challenge？

chensong1995 commented 1 year ago

Hi weidu3,

The dataset is downloaded from this link: https://hci.iwr.uni-heidelberg.de/vislearn/iccv2015-occlusion-challenge/

As stated in the link: "NOTE: Below you find the version of the occlusion dataset as it was used in our ICCV15 challenge. However, we released a reworked version of the dataset as part of the BOP Challenge. The reworked version contains all data (images, poses, 3D models of objects) and some annotation errors have been corrected. We advise to use the reworked version of the dataset.".

The "reworked version" doesn't load correctly for me (503 Service Unavailable). The experiments use the original version from the ICCV15 challenge.

I hope this clarifies! Let me know if you have further concerns.