search

open-mmlab / mmdetection

OpenMMLab Detection Toolbox and Benchmark
https://mmdetection.readthedocs.io
Apache License 2.0
29.21k stars 9.4k forks source link

BUG in evaluating lvis proposals #10606

Open fushh opened 1 year ago

fushh commented 1 year ago

https://github.com/open-mmlab/mmdetection/blob/f78af7785ada87f1ced75a2313746e4ba3149760/mmdet/evaluation/metrics/lvis_metric.py#L295-L296

When evaluating proposals, we ignore the categories. However, the variables of params in lvis eval is not the same as coco eval. It should be lvis_eval.params.use_cats = 0 and lvis_eval.params.max_dets = list(self.proposal_nums).

Please check the source code and fix the bug.

yanjk3 commented 1 year ago

+1

hhaAndroid commented 1 year ago

@fushh Thank you.