follow-me

You can run this from a folder with images (which correspond to video frames), or from a video. Also, you can skip detection phase by providing a detection file (MOT format).

Create an ENV variable with the code path

export CODE_PATH=/Volumes/128GB/Maestria

Run from images

python3 run.py --images_path=$CODE_PATH/Datasets/MOT16/test/MOT16-06/img1

Run from video

python3 run.py --video_input=$CODE_PATH/Datasets/test.mp4

Do not run detection phase and use detections from file

python3 run.py --images_path=$CODE_PATH/Datasets/MOT16/test/MOT16-06/img1 --load_detection_file=$CODE_PATH/Datasets/MOT16/test/MOT16-06/det/det.txt

Run deep sort

python deep_sort_app.py --sequence_dir=../../../Datasets/MOT16/test/MOT16-06 --detection_file=$CODE_PATH/Datasets/MOT16/resources_deep_sort/detections/MOT16_POI_test/MOT16-06.npy --min_confidence=0.3 --nn_budget=100 --display=True

Remove useless files (causes conflics when reading images from folder)

cd $CODE_PATH/Datasets/MOT16/train
find . -type f -name '._*' -delete

Run metrics (performance, accuracy) - Works with deep sort for now

Deep sort

Create tracking files

cd $CODE_PATH/Code/lib/deep_sort
python evaluate_motchallenge.py --mot_dir=$CODE_PATH/Datasets/MOT16/train --detection_dir=$CODE_PATH/Datasets/MOT16/resources_deep_sort/detections/MOT16_POI_train

Evaluate performance Uses: https://github.com/cheind/py-motmetrics

python -m motmetrics.apps.eval_motchallenge $CODE_PATH/Datasets/MOT16/train $CODE_PATH/Code/lib/deep_sort/results --fmt mot16

Metrics using deep_sort MOT detections: 11:09:21 INFO - Found 7 groundtruths and 7 test files. 11:09:21 INFO - Available LAP solvers ['lap', 'scipy'] 11:09:21 INFO - Default LAP solver 'lap' 11:09:21 INFO - Loading files. 11:09:25 INFO - Comparing MOT16-13... 11:09:32 INFO - Comparing MOT16-09... 11:09:35 INFO - Comparing MOT16-10... 11:09:41 INFO - Comparing MOT16-11... 11:09:46 INFO - Comparing MOT16-05... 11:09:49 INFO - Comparing MOT16-02... 11:09:58 INFO - Comparing MOT16-04... 11:10:32 INFO - Running metrics IDF1 IDP IDR Rcll Prcn GT MT PT ML FP FN IDs FM MOTA MOTP MOT16-13 56.1% 52.9% 59.8% 80.3% 71.1% 107 65 36 6 3745 2250 313 352 44.9% 0.237 MOT16-09 56.2% 64.3% 49.9% 71.5% 92.1% 25 13 11 1 322 1497 41 59 64.6% 0.161 MOT16-10 54.0% 53.4% 54.6% 76.4% 74.8% 54 24 29 1 3178 2905 242 310 48.7% 0.228 MOT16-11 63.3% 68.3% 58.9% 76.2% 88.4% 69 29 33 7 918 2187 65 96 65.4% 0.152 MOT16-05 60.8% 72.6% 52.3% 62.3% 86.4% 125 29 69 27 667 2573 65 115 51.5% 0.215 MOT16-02 40.7% 46.4% 36.2% 53.9% 69.1% 54 12 31 11 4302 8215 138 252 29.0% 0.210 MOT16-04 70.0% 76.0% 64.8% 71.9% 84.3% 83 42 26 15 6358 13358 71 255 58.4% 0.167 OVERALL 60.2% 64.4% 56.6% 70.1% 79.9% 517 214 235 68 19490 32985 935 1439 51.6% 0.189 11:23:50 INFO - Completed

Follow me

python3 run.py --images_path=/Volumes/128GB/Maestria/Datasets/MOT16/train/MOT16-02/img1 --store_detections=results/MOT16-02.txt --no-display

python -m motmetrics.apps.eval_motchallenge $CODE_PATH/Datasets/MOT16/train2 /Volumes/128GB/Maestria/Code/results --fmt mot16 02:26:17 INFO - Found 1 groundtruths and 1 test files. 02:26:17 INFO - Available LAP solvers ['lap', 'scipy'] 02:26:17 INFO - Default LAP solver 'lap' 02:26:17 INFO - Loading files. 02:26:17 INFO - Comparing MOT16-02... 02:26:24 INFO - Running metrics IDF1 IDP IDR Rcll Prcn GT MT PT ML FP FN IDs FM MOTA MOTP MOT16-02 27.1% 42.0% 20.0% 28.9% 60.6% 54 7 19 28 3343 12683 88 170 9.6% 0.269 OVERALL 27.1% 42.0% 20.0% 28.9% 60.6% 54 7 19 28 3343 12683 88 170 9.6% 0.269 02:26:40 INFO - Completed