Clarification on Table-7

[sandeepjana]

Thank you for your work. In paper "Table 7: Metrics for segmentation masks*, are the AP scores AP_mask or AP_bbox? Though AP_mask seems natural metric here, I am confused because "Fig. 5: Mask R-CNN training" mentions about AP_bbox metrics

[jdkuhnke]

I assume you are referring to this paper on arXiv.

Table 6 shows the AP for the bounding boxes and Table 7 shows the AP for the instance segmentation masks.

For a short discussion about appropriate metrics see "3.5.2 AP and IoU for evaluation in practical applications" from my thesis: relevant excerpt.

Cheers,
Dominik