search

DecaYale / RNNPose

RNNPose: Recurrent 6-DoF Object Pose Refinement with Robust Correspondence Field Estimation and Pose Optimization, CVPR 2022
Apache License 2.0
160 stars 17 forks source link

A problem about the results. #9

Closed pyni closed 2 years ago

pyni commented 2 years ago

Hi, Thanks for your wonderful works. I have tried your codes, but It seems the test result doesn't match with the result in your paper. I just follow all the steps from your webside and the config file is the copy from template_fw0.5.yml. Then I run "bash scripts/eval.sh" and I test two objects (ape and cat). The result is shown as follows:

For ape: 2d projections metric: 0.9857142857142858 ADD metric: 0.780952380952381 ADD2 metric: 0.1380952380952381 ADD5 metric: 0.4219047619047619 5 cm 5 degree metric: 0.9876190476190476 mask ap70: nan seq_len: 1050

For cat: ADD metric: 0.9421157684630739 ADD2 metric: 0.2275449101796407 ADD5 metric: 0.6117764471057884 5 cm 5 degree metric: 0.9940119760479041 mask ap70: nan seq_len: 1002

I cannot find the results in your paper I don't know why?

DecaYale commented 2 years ago

Hi, do you install the environment with the docker file provided? We have tested locally and the results should match or are at least close to those reported in our paper.

pyni commented 2 years ago

OK, I install it using miniconda. I will try docker later.

pyni commented 2 years ago

Hi, may I ask another question? the runtime analysis in table 2 is for one iteration or five iterations? It seems not clear. Thanks.

DecaYale commented 2 years ago

Table 2 provides the runtime of each individual module per forwarding.

pyni commented 2 years ago

OK, I test it in my PC with 0.7179 second per image using GeForce GTX 1660 Ti.

Arrebol2020 commented 1 year ago

OK, I install it using miniconda. I will try docker later.

Hi，Did you get the same result with the ape model. For ape, I also get 78.1.