The official resitory for 8th NVIDIA AI City Challenge (Track1: Multi-Camera People Tracking) from team NetsPresso (Nota Inc.).
# git clone this repository
git clone https://github.com/nota-github/AIC2024_Track1_Nota.git
cd AIC2024_Track1_Nota
The official dataset is available for download at https://www.aicitychallenge.org/2024-data-and-evaluation/, the website of the AI City Challenge.
To get the password to download them, you must complete the dataset request form.
(We are not permitted to share the dataset, per the DATASET LICENSE AGREEMENT from the dataset author(s).)
AIC2024_Track1_Nota
└── data
└── videos
├── train
│ ├── scene_001
│ │ ├── camera_0001
│ │ │ ├── calibration.json
│ │ │ └── video.mp4
│ │ ├── ...
│ │ └── ground_truth.txt
│ ├── scene_002
│ ├── ...
├── val
│ ├── ...
└── test
├── ...
bash scripts/generate_all_datasets.sh
bash scripts/generate_only_frames.sh
# Build a docker image
docker build -t aic2024/track1_nota:latest .
# Build a docker container
docker run -it --gpus all --shm-size=8g \
-v /path/to/AIC2024_Track1_Nota:/home/workspace/AIC2024_Track1_Nota \
-v /path/to/AIC2024_Track1_Nota/data:/workspace/ \
aic2024/track1_nota:latest /bin/bash
Train People Detection Model
bash scripts/train_od.sh
Train ReID Model
bash scripts/train_reid.sh
If you want to use pretrained models, please download them from the provided Google Drive and place them in the './pretrained' directory.
Option1: Inference each scene sequentially
bash scripts/run_mcpt.sh
Option2: Inference scenes in parallel (to get a faster results)
bash scripts/run_mcpt_parallel.sh
(If errors occur, inference only on the affected scenes separately, then run 'python3 tools/merge_results.py')
The result files will be saved as follows:
AIC2024_Track1_Nota
└── results
├── scene_061.txt
├── ...
└── track1_submission.txt