Mirrors/blakeblackshear.frigate
1
0
Fork 0
mirror of https://github.com/blakeblackshear/frigate.git synced 2024-11-21 19:07:46 +01:00
Code Issues Projects Releases Wiki Activity

detectors/yolo_utils: use nms to prefilter overlapping boxes if too many detected

Browse Source
This commit is contained in:
Indrek Mandre 2024-02-02 16:54:39 +02:00
parent 61713115e2
commit cd508980bb
1 changed files with 15 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
frigate/detectors/yolo_utils.py
View File

@ -13,20 +13,31 @@ def yolov8_preprocess(tensor_input, model_input_shape):
# cv2.dnn.blobFromImage is faster than numpying it # cv2.dnn.blobFromImage is faster than numpying it
return cv2.dnn.blobFromImage(tensor_input[0], 1.0 / 255, (model_input_shape[3], model_input_shape[2]), None, swapRB=False) return cv2.dnn.blobFromImage(tensor_input[0], 1.0 / 255, (model_input_shape[3], model_input_shape[2]), None, swapRB=False)
def yolov8_postprocess(model_input_shape, tensor_output, box_count = 20): def yolov8_postprocess(model_input_shape, tensor_output, box_count = 20, score_threshold = 0.3, nms_threshold = 0.5):
model_box_count = tensor_output.shape[2] model_box_count = tensor_output.shape[2]
probs = tensor_output[0, 4:, :] probs = tensor_output[0, 4:, :]
all_ids = np.argmax(probs, axis=0) all_ids = np.argmax(probs, axis=0)
all_confidences = probs.T[np.arange(model_box_count), all_ids] all_confidences = probs.T[np.arange(model_box_count), all_ids]
all_boxes = tensor_output[0, 0:4, :].T all_boxes = tensor_output[0, 0:4, :].T
mask = (all_confidences > 0.30) mask = (all_confidences > score_threshold)
class_ids = all_ids[mask] class_ids = all_ids[mask]
confidences = all_confidences[mask] confidences = all_confidences[mask]
cx, cy, w, h = all_boxes[mask].T cx, cy, w, h = all_boxes[mask].T
scale_y, scale_x = 1 / model_input_shape[2], 1 / model_input_shape[3]
if model_input_shape[3] == 3:
scale_y, scale_x = 1 / model_input_shape[1], 1 / model_input_shape[2]
else:
scale_y, scale_x = 1 / model_input_shape[2], 1 / model_input_shape[3]
detections = np.stack((class_ids, confidences, scale_y * (cy - h / 2), scale_x * (cx - w / 2), scale_y * (cy + h / 2), scale_x * (cx + w / 2)), axis=1) detections = np.stack((class_ids, confidences, scale_y * (cy - h / 2), scale_x * (cx - w / 2), scale_y * (cy + h / 2), scale_x * (cx + w / 2)), axis=1)
if detections.shape[0] > box_count: if detections.shape[0] > box_count:
detections = detections[np.argpartition(detections[:,1], -box_count)[-box_count:]] # if too many detections, do nms filtering to suppress overlapping boxes
boxes = np.stack((cx - w / 2, cy - h / 2, w, h), axis=1)
indexes = cv2.dnn.NMSBoxes(boxes, confidences, score_threshold, nms_threshold)
detections = detections[indexes]
# if still too many, trim the rest by confidence
if detections.shape[0] > box_count:
detections = detections[np.argpartition(detections[:,1], -box_count)[-box_count:]]
detections = detections.copy()
detections.resize((box_count, 6)) detections.resize((box_count, 6))
return detections return detections