how can i use anthor reid model in this work

[szh91]

hi, thank you for the wonderful work and offer the code for us. If I train a separate ReID model, then I want to add this model in the code, how I can do this?

[selflein]

Hi,
it should work pretty much out of the box by replacing the net variable here with your pretrained ReID model.

GraphNN-Multi-Object-Tracking/src/data_utils/preprocessing.py

Lines 201 to 206 in 74bf969

	net = osnet_x0_5(pretrained=True).to(args.device)
	net.eval()
	ds = MOT16(args.dataset_path, args.mode, vis_threshold=args.threshold)
	pca = pickle.load(open(args.pca_path, 'rb'))
	preprocess(output_dir, net, ds, pca, args.save_imgs, device=args.device)

Note that this script uses PCA dimensionality reduction to reduce the feature size to 32 and is tuned for the ReID model used. So you may have to obtain a new PCA via

GraphNN-Multi-Object-Tracking/src/data_utils/preprocessing.py

Line 49 in 74bf969

def fit_pca(save_path: str, dataset_path: str, re_id_net):

if feature sizes of your ReID method are not maching or remove the PCA entirely and add a MLP to the main model reducing the feature size.