Open eeric opened 3 years ago
Hi eeric! I may have the same question with you. Is this iou-tracker code tracks frame by frame? I feel it does iou tracking by loading all frames detections.
My solution is loading my detection result every "mint" frames, then get tracks of "mint" frames.
load_dection = []
mint = 2
while(True):
# read a frame
id = 0
for c in contours:
(x, y, w, h) = cv2.boundingRect(c)
load_dection = np.concatenate((load_dection, np.array([[cap.get(cv2.CAP_PROP_POS_FRAMES), id, x, y, w, h, 1]])), axis = 0)
id += 1
minT += 1
if minT == mint:
detections = load_mot(load_dection, nms_overlap_thresh=None, with_classes=False)
tracks = track_iou(detections, 0, 0.5, 0.5, mint)
load_dection = []
minT = 0
I have the same question too!!!
@DataXujing @tkamkb any lead on this?
thanks