Reported results for SCG

[daoyuan98]

Hi, Thank you for the great work!

In this paper, the reported results for SCG on HICO-DET is 29.26 mAP (full), while the reported result on SCG paper was 31.33 mAP (full). Is the difference caused by different backbones used (i.e. ResNet50-FPN v.s. ResNet101)? Thank you!

[fredzzhang]

Hi @daoyuan98,

The 31.33 mAP was obtained using a set of fine-tuned detections provided by the DRG repo, which was later found to be not reproducible, as noted here. With detections we can reproduce, as noted in the paper, SCG has an mAP of 24.88 using fine-tuned Faster R-CNN, and 29.26 using fine-tuned DETR.

Hope that clarifies things.

Fred.

[daoyuan98]

Thank you so much for your information!