search

j96w / 6-PACK

"6-PACK: Category-level 6D Pose Tracker with Anchor-Based Keypoints" code repository
https://sites.google.com/view/6packtracking
MIT License
297 stars 52 forks source link

Your evaluation perhaps has some problem? #12

Closed Guptajakala closed 4 years ago

Guptajakala commented 4 years ago

https://github.com/j96w/6-PACK/blob/cac1b29f3e7a5e9532de610b4f5516919edb0063/dataset/eval_list/eval_list_1.txt#L5 Here the "bottle_shengjun_norm scene_3" appeared twice. That might affect your final result on the bottle category.

j96w commented 4 years ago

Hi, please refer to the benchmark.py to do the evaluation, and in this code, we didn't use the eval_list.txt. The only problem of this duplicate is during the first step of our evaluation, where we use the model to acquire the pose prediction results, the model will test this object twice and the second result will cover the previous one. In the second step, the benchmark.py will get a score from the pose prediction results, which won't count the score of this object twice since the first one has already been covered by the second one.

Guptajakala commented 4 years ago

This answered