search

xingyizhou / CenterTrack

Simultaneous object detection and tracking using center points.
MIT License
2.39k stars 526 forks source link

MOT17 Train Custom dataset with static images or sequential data? #259

Open MichiTrain opened 2 years ago

MichiTrain commented 2 years ago

Does the command for training MOT17 as custom dataset uses static images for training? Or does it train the model with video data/sequential data?

python main.py tracking --exp_id mot17_half_sc --dataset custom --custom_dataset_ann_path ../data/mot17/annotations/train_half.json --custom_dataset_img_path ../data/mot17/train/ --input_h 544 --input_w 960 --num_classes 1 --pre_hm --ltrb_amodal --same_aug --hm_disturb 0.05 --lost_disturb 0.4 --fp_disturb 0.1 --gpus 0,1

I am not sure, which kinds of arguments define, if the model uses static or video data. I would assume, that static data are only used, when the dataset as COCO defined is:--dataset coco And when --dataset custom is used, then the training uses video data/sequential data. I am not sure, if this is true.