How to run inference on videos directly

[usman-bsee]

Dear Author,
Thanks for your interest, Please can you guide how to run inference on new videos without loading datasets in the dataset directory.
Thanks.

[ret-1]

Hello,

In inference, the get_eval_loader function defined in the exp file is used to get the DataLoader:

MixSort/tools/track_mixsort.py

Lines 175 to 183 in a078f5b

	val_loader = exp.get_eval_loader(args.batch_size, is_distributed, args.test, return_origin_img=True)
	evaluator = MOTEvaluator(
	args=args,
	dataloader=val_loader,
	img_size=exp.test_size,
	confthre=exp.test_conf,
	nmsthre=exp.nmsthre,
	num_classes=exp.num_classes,
	)

Since we haven't provided a script to directly run inference on custom videos, you need to manually convert it to coco format, create a new exp file, and modify the get_eval_loader function to correctly read your data.

You could refer to yolox_x_sportsmot.py and convert_sportsmot_to_coco.py