AIC20 Track 1:Vehicle Counts by Class at Multiple Intersections

Overview

Our implementation comprised of:

1/ Detected bounding box based on dla-centernet architecture

2/ Tracking car moving based on IoU overlapping

3/ Counting car based on densly sliding object through MOI region

Usage

Install Detector - CenterNet

If you are familiar with Vietnamese, you can read 'CenterNet/SETUP.md' for more information.

Default GPU

I ran this code with NVIDIA Tesla K80. If you face a problem about GPU version, you can config GPU_version in CenterNet resources.

Download Dataset

Download images-set: https://drive.google.com/open?id=1xFcfOEfAXjjzdrZbH3glOxv0rHIne8H7
Download labels: https://drive.google.com/file/d/1SWsjrSNaRp3CVe9h3Fu41ezkvXcGPy0_
Create folders: 'abc', 'abc/images' and 'abc/labels' into 'data' folder.
Unzip images-set, then move all images from images-set folder to 'data/abc/images'.
Unzip labels, then move all json files to 'data/abc/labels'.

Install COCOAPI

cd AIC20/track1-multi-intersection-counting
COCOAPI = 'cocoapi'
git clone https://github.com/cocodataset/cocoapi.git 'cocoapi'
cd $COCOAPI/PythonAPI
make
python setup.py install --user

Install CenterNet

cd AIC20/track1-multi-intersection-counting
CenterNet_ROOT = 'CenterNet'
cd $CenterNet_ROOT
pip install -r requirements.txt
cd $CenterNet_ROOT/src/lib/external
python setup.py build_ext --inplace

Install DCN2

cd $CenterNet_ROOT/src/lib/models/networks/DCNv2
python setup.py build develop

Pretrained Model

https://www.dropbox.com/s/q9jimptc5e8e2we/model_best_dla_1x.pth?dl=0

Training (or you can use pretrained model and skip this step)

cd $CenterNet_ROOT/src
python main.py ctdet --exp_id abc_dla_34 --arch dla_34 --batch_size 32 --num_workers 4 --num_epochs 100

Remember: after training, all models are saving in 'exp/ctdet/<exp_id>/model_best.pth'

Inference

Firstly, you must have a pretrained model. Then, config my code:

cd $CenterNet_ROOT/src
python demo_video.py ctdet --arch dla_34 --load_model <link to model> --demo <link to test-dataset>

Remember: after inference, your bounding box results are saving in 'CenterNet/Detection/bboxes_<video_name>'

Reproduce tracking

cd AIC20

python test_iou.py

Reproduce counting

cd AIC20/car_counter

python counter.py

Acknowledgement

Source code for tracking car is built based on iou tracking of High-Speed Tracking-by-Detection Without Using Image Information

nttung1110/AIC20