Latest Yolov4 Models Not Working

[rogerquake]

I have been trying to load in the latest Yolov4 models below to Viseron for testing and when I do, no objects get detected.

https://github.com/AlexeyAB/darknet/releases/tag/yolov4

The scaled Yolov4 model seems interesting, so was trying to see if that would work.

[roflcoopter]

Great idea, hadn't seen those!
I need to upgrade OpenCV for it to work, will do so soon!

[rogerquake]

That would be awesome, thanks! I updated the OpenCV version to the latest release, 4.5.4 in the .env file and compiled the OpenCV docker container from source, but it still didn't work.

[roflcoopter]

Did you also build the main image? They are chained together

[rogerquake]

Yeah, built the OpenCV image followed by the main image. From reading the github threads for OpenCV, it's hard to tell if the required updates are actually in the latest build or not. opencv/opencv#18975

[roflcoopter]

Seems to be in 4.5.4 according to this comment

[rogerquake]

I see this error in the debug log when it attempts to perform object detection with the yolov4-csp weight and cfg:

viseron | Exception in thread object_detection:
viseron | Traceback (most recent call last):
viseron | File "/usr/lib/python3.8/threading.py", line 932, in _bootstrap_inner
viseron | self.run()
viseron | File "/usr/lib/python3.8/threading.py", line 870, in run
viseron | self._target(*self._args, **self._kwargs)
viseron | File "/src/viseron/detector/init.py", line 105, in object_detection
viseron | frame_to_scan.frame.objects = self.object_detector.return_objects(
viseron | File "/src/viseron/detector/darknet/init.py", line 108, in return_objects
viseron | objects = self.post_process(labels, confidences, boxes)
viseron | File "/src/viseron/detector/darknet/init.py", line 81, in post_process
viseron | self.labels[int(label[0])],
viseron | IndexError: invalid index to scalar variable.

[roflcoopter]

Seems the code has to change as well, see here

Not sure how to apply it right now tho but i will get to it soon!

[ozett]

let us know if when can try yolo4...

[roflcoopter]

Will be in #306 when its done

[ozett]

i will follow the progess on #306
thanks for notifying here 🤝

[roflcoopter]

YOLOv4 and even YOLOv7 is available in v2