I’ve written Python code to use “ssd-mobilenet -v2” uff model with “jetson.inference.detectNet” to perform inference on video data.
Inference is working well, but the bounding box co-ordinates for detected objects seems to be incorrect.
In the attached image, it seems that the bounding boxes needs to be “transposed” vertically.
Below is the sample code used:
width = frame.shape height = frame.shape img = jetson.utils.cudaFromNumpy(frame) detections = net.Detect(img, width, height, overlay='none') for detection in detections: detected_class = class_names[detection.ClassID] left = int(detection.Left) right = int(detection.Right) width = int(detection.Width) height = int(detection.Height) top = int(detection.Top) bottom = int(detection.Bottom) cv2.rectangle(frame, (top, left), (bottom, right), (0,255,0), 2) cv2.putText(frame, detected_class, (top-10, left-10), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0,255,0), 1)
Can somebody help me with this ?