YOLOv9 + DeepSORT

[sujanshresstha]

Hi,I hope you're doing well. I've integrated YOLOv9 with DeepSORT in the following repository: https://github.com/sujanshresstha/YOLOv9_DeepSORT.git

[jdiaz97]

Oh, no. No more AI-generaated messages, please.

[jdiaz97]

Congrats @sujanshresstha seems very good!

[sujanshresstha]

Hi @jdiaz97,
Thank you for your positive feedback! Indeed, integrating YOLOv9 with DeepSORT opens up a lot of exciting possibilities. I appreciate your support!

[WongKinYiu]

Added to readme.

[WongKinYiu]

@sujanshresstha

I am not sure, but you may need to check which one is correct.
for det in results.pred[0]: or for det in results.pred[1]:
https://github.com/sujanshresstha/YOLOv9_DeepSORT/blob/main/object_tracking.py#L60

[sujanshresstha]

@WongKinYiu,

I have opted to use for det in results.pred[0]: instead of for det in results.pred[1]: due to the following reason:

Inside the DetectMultiBackend class, there are lines of code:

if self.pt:  # PyTorch
    y = self.model(im, augment=augment, visualize=visualize) if augment or visualize else self.model(im)

if isinstance(y, (list, tuple)):
    return self.from_numpy(y[0]) if len(y) == 1 else [self.from_numpy(x) for x in y]

Since the model outputs an aggregated single tensor even for multiple object detection, len(y) == 1.

I have double-checked with for det in results.pred[1]:, but it throws an IndexError: list index out of range since len(results.pred) = 1.

[karthikbalu]

Bytetrack is better for practical application than Deepsort for less latency