How can I fix ValueError in object detection application?

Question

I am trying to add bounding boxes to an image using onnxruntime and opencv to be detect objects with the yolov2 neural network. Instead, I get an error at runtime.

I converted the input image into a compatible tensor / numpy array to feed into the model. Once I knew everything worked perfectly without bugs, I added the following code to add the bounding boxes:

    while True:
        gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        for x, y, w, h in pred_onnx:
            cv2.rectangle(img, (x, y), (x+w, y+h), (0, 255, 0), 2)
            roiGray = gray[y:y+h, x:x+w]
            roiColor = img[y:y+h, x:x+w]

        cv2.imshow("Detect", cv2.resize(img, (500, 500)))
        cv2.waitKey(0)

I was expecting the image to show (green) bounding boxes. Instead, I get this error:

File "C:\Users\MyName\Desktop\OnnxCV\onnxcv\object_detector.py", line 27, in <module>
    for x, y, w, h in pred_onnx:
ValueError: not enough values to unpack (expected 4, got 1)

The full code is here if it helps.

Answer 1

The pred_onnx array is not in the shape expected by the current code--there is some more postprocessing to do. See here for details about the output.

For example, using the 30% threshold suggested by the linked post, you would loop through and filter the bounding boxes like so:

for r in range(13):
    for c in range(13):
        confidence = pred_onnx[0, 4, r, c]
        if confidence < 0.3:
            continue
        x = pred_onnx[0, 0, r, c]
        y = pred_onnx[0, 1, r, c]
        w = pred_onnx[0, 2, r, c]
        h = pred_onnx[0, 3, r, c]
        ...