This model is an instance segmentation network for 80 classes of objects. It is a Mask R-CNN with ResNet50 backbone, FPN, RPN, detection and segmentation heads.
|COCO val2017 box AP (max short side 768, max long side 1024)||40.8%|
|COCO val2017 mask AP (max short side 768, max long side 1024)||36.9%|
|COCO val2017 box AP (max height 768, max width 1024)||39.86%|
|COCO val2017 mask AP (max height 768, max width 1024)||36.44%|
|Max objects to detect||100|
Average Precision (AP) is defined and measured according to standard COCO evaluation procedure.
1, 3, 768, 1024 in the format
1, C, H, W, where:
C- number of channels
H- image height
W- image width
The expected channel order is
100- Contiguous integer class ID for every detected object.
100, 5- Bounding boxes around every detected objects in (top_left_x, top_left_y, bottom_right_x, bottom_right_y) format and its confidence score in range [0, 1].
100, 28, 28- Segmentation heatmaps for every output bounding box.
[*] Other names and brands may be claimed as the property of others.