BBoxSelector

class savant.selector.detector.BBoxSelector(confidence_threshold=0.5, nms_iou_threshold=0.5, min_width=0, min_height=0, max_width=0, max_height=0, **kwargs)

Detector bbox per class selector.

Parameters:

confidence_threshold (float) – confidence threshold
nms_iou_threshold (float) – nms iou threshold
min_width (int) – minimal bbox width
min_height (int) – minimal bbox height
max_width (int) – maximum bbox width
max_height (int) – maximum bbox height

__call__(bbox_tensor)

Filters bboxes by confidence and size, applies NMS.

Parameters:: bbox_tensor (ndarray) – tensor(class_id, confidence, left, top, width, height)
Returns:: Selected bbox tensor
Return type:: ndarray