open-mmlab
regarding MoT2reid
I am experimenting with MMTracking and I am currently processing the MOT20 dataset(full version) Although I have been able to create the annotations and pickle files using
python ./tools/convert_datasets/mot/mot2coco.py -i ./data/MOT17/ -o ./data/MOT17/annotations --split-train --convert-det
It seems that the execution has stalled when i run the comand
python ./tools/convert_datasets/mot/mot2reid.py -i ./data/MOT20/ -o ./data/MOT20/reid --val-split 0.2 --vis-threshold 0.3
the console output is as follows
Converting train set to COCO format
100% 4/4 [00:13<00:00, 3.41s/it]
train has 2215 instances.
Done! Saved as ./data/MOT20/annotations/train_cocoformat.json and ./data/MOT20/annotations/train_detections.pkl
Converting test set to COCO format
100% 4/4 [00:01<00:00, 3.11it/s]
test has 0 instances.
Done! Saved as ./data/MOT20/annotations/test_cocoformat.json and ./data/MOT20/annotations/test_detections.pkl
Converting half-train set to COCO format
100% 4/4 [00:36<00:00, 9.17s/it]
half-train has 1241 instances.
Done! Saved as ./data/MOT20/annotations/half-train_cocoformat.json and ./data/MOT20/annotations/half-train_detections.pkl
Converting half-val set to COCO format
100% 4/4 [00:37<00:00, 9.34s/it]
half-val has 1418 instances.
Done! Saved as ./data/MOT20/annotations/half-val_cocoformat.json and ./data/MOT20/annotations/half-val_detections.pkl
0% 0/4 [00:00<?, ?it/s]y
when I click on the console,it seems I am being asked to enter some prompt as indicated by the presence of the y.Are we required to enter something in the prompt.Plus how much time does it take for the entire dataset to be processed.
Secondly when I executed the sample demo script the result video is only 2 seconds,is there a way to increase the duration of the video so that the full tracking results can be observed clearly. I am currently running the code on colab and the MMCV,MMDet,pytorch,cuda and MMtrack version are listed below
Pytorch 1.11.0+cu113 True
Cuda 11.1
GCC 7.3
MMDet 2.23.0
MMtrack 0.12.0
Seems that batch 1/4 has been completed.It took almost 2 hours to process.Is that consistent.also,what can be done to improve the conversion time.
With using Intel(R) Xeon(R) Gold 6348 CPU @ 2.60GHz CPU, it took me around 2 hours to finish the conversion, which is acceptable to me. If you get multiple CPUs, I would recommend to try multiple processes (not multi-threading) to accelerate the process.
