search

thodan / bop_toolkit

A Python toolkit of the BOP benchmark for 6D object pose estimation.
http://bop.felk.cvut.cz
MIT License
397 stars 138 forks source link

Is the SiSo task suitable for Bop challenge? #35

Closed TouchDeeper closed 4 years ago

TouchDeeper commented 4 years ago

Hello,

My method focuses on pose registration of object pose estimation, object segmentation is not yet developed. Now we can solve the SiSo problem, Is the SiSo task suitable for Bop challenge? Thanks a lot!

thodan commented 4 years ago

Hello, the BOP Challenge (bop.felk.cvut.cz) is focused on the ViVo task (i.e. a generalization of the SiSo task). The ECCV'18 paper (http://cmp.felk.cvut.cz/~hodanto2/data/hodan2018bop.pdf) considers the SiSo task and you can compare your method to the methods in the paper.

TouchDeeper commented 4 years ago

@thodan, I want to ask further whether the input of ECCV'18 paper is the depth contained in mask_visib or just the mask.

thodan commented 4 years ago

@TouchDeeper, the masks in the "mask" and "mask_visib" are supposed to be used only for training and for the calculation of the object's visibility. The masks may not be used at the test time -- the test input of the methods evaluated in the ECCV'18 paper was only a single RGB-D image + identifiers of the objects present in the image.

TouchDeeper commented 4 years ago

@thodan I didn't develop the object identifiers. My method requires the point cloud of the instance (what the mask can do) and the object to which the instance belongs. Could you please provide some advice on how to use your bop toolkit to evaluate my method?

thodan commented 4 years ago

@TouchDeeper, you can prepare your 6D pose estimates in the BOP Challenge format (https://bop.felk.cvut.cz/challenges/bop-challenge-2020/#howtoparticipate) and submit to the BOP evaluation system. To get the masks at test time, you may use e.g. Mask R-CNN.

TouchDeeper commented 4 years ago

@thodan Thanks for your advice! I will try it.