coldmanck / VidHOI

Official implementation of "ST-HOI: A Spatial-Temporal Baseline for Human-Object Interaction Detection in Videos" (ACM ICMRW 2021)

https://dl.acm.org/doi/10.1145/3463944.3469097

Bug in evaluation?

zyong812 opened this issue 2 years ago · comments

Yong Z commented 2 years ago

in file https://github.com/coldmanck/VidHOI/blob/master/vidor_eval.ipynb.

We should not just skip these triplets. Instead, ap for these triplets should be set as 0.

Meng-Jiun Chiou commented 2 years ago

Good catch, that makes sense. Please feel free to add a PR and report the revised numbers.