This is a person detector that is based on Cascade R-CNN architecture with ResNet50 backbone.
Metric | Value |
---|---|
AP | 94.54% (internal test set) |
Pose coverage | Standing upright, parallel to image plane |
Support of occluded persons | YES |
Occlusion coverage | <50% |
Min person height | 100 pixels (on 1080p) |
GFlops | 404.264 |
MParams | 71.565 |
Source framework | PyTorch* |
Average Precision (AP) is defined as an area under the precision/recall curve. Intersection over union threshold of 0.5 is used for matching.
Name: input
, shape: [1x3x800x1344] - An input image in the format [BxCxHxW], where:
Expected color order is BGR.
boxes
is a blob with the shape [N, 5], where N is the number of detected bounding boxes. For each detection, the description has the format [x_min
, y_min
, x_max
, y_max
, conf
], where:x_min
, y_min
) - coordinates of the top left bounding box cornerx_max
, y_max
) - coordinates of the bottom right bounding box cornerconf
- confidence for the predicted classlabels
is a blob with the shape [N], where N is the number of detected bounding boxes. It contains label
per each detected box.[*] Other names and brands may be claimed as the property of others.