search

hustvl / SparseTrack

Official PyTorch implementation of SparseTrack (the new version of code will come soon)
MIT License
135 stars 12 forks source link

Using other Object Detectors #4

Closed calvpang closed 1 year ago

calvpang commented 1 year ago

From what I understand, the model training is specifically to train an object detector with YOLOX and not training the tracker itself. If I had a functioning YOLOv5 model, I'm assuming I could use torch.load() and run inference on each frame using that. But I'm a bit lost on where I can do this with the current code.

Xavier-Lin commented 1 year ago

Thank you for your attention. We will release the inference version of YOLOV8.

Mittag-tech commented 1 year ago

Hi @Xavier-Lin , I have almost the same question. Can I change to another detector by myself?

If the model training is to train a detector as @calvpang says, I think I can use it by setting config to a path of weghts created by training a detector separately. Is this correct?

Xavier-Lin commented 1 year ago

Hi @Mittag-tech，actually, there's no problem with this. In general, you can achieve the goal of changing the detector for training by replacing the detector model in the model folder and modifying the corresponding detector settings in the xxx_config.py file.

Mittag-tech commented 1 year ago

Thank you for your response, @Xavier-Lin . I would try it.

Just to confirm, is this correct that the parameters of tracking module by pseudo-depth are adjusted manually?

Xavier-Lin commented 1 year ago

@Mittag-tech Manually adjusting the number of layers for pseudo-depth is possible, if needed. In general, we do not adjust the number of layers for pseudo-depth, but we do adjust the matching threshold.

Mittag-tech commented 1 year ago

Thank you for your information, @Xavier-Lin !