AIC21-track1-team19

Introduction

In this repo, we include the submission to AICity Challenge 2021 Vehicle Counts by Class at Multiple Intersections.

Our implementation comprised of:

(1) we re-designed a detection-tracking-counting (DTC) for movement-specific vehicle counting problem regard to both effectiveness and efficiency.

(2) We modified Deep SORT with the efficient features to improve the multiple objects tracking performance.

(3) We proposed the cosine similarity-based and orbit-based nearest neighbor analysis to improve the vehicle counting performance.

Detector

Download Dataset

1. Download images-set and lables : link

2. Unzip data, then move all to 'AIC21-track1-team19/ScaledYOLOv4-yolov4-csp/data/'.

install mish-cuda

cd mish-cuda-master python setup.py build install

For install environment:

pip install -r requirements.txt

weights

1. Download weights : link

2. Unzip weights, then move all to 'AIC21-track1-team19/ScaledYOLOv4-yolov4-csp/'.

Trainning

Go to 'ScaledYOLOv4-yolov4-csp'

you can change batch size to fit your GPU RAM.

python train.py --device 0 --batch-size 8 --data ./data/data.yaml --cfg yolov4-csp-3-0.25.cfg --name yolov4-csp-3-0.25 --hyp ./data/hyp.finetune.yaml --img-size 512 512 --weight yolov4-csp.weights --epoch 300

For resume training: assume the checkpoint is stored in runs/exp0_yolov4-csp-3-0.25/weights/.

python train.py --device 0 --batch-size 8 --data ./data/data.yaml --cfg yolov4-csp-3-0.25.cfg --weights 'runs/exp0_yolov4-csp-3-0.25/weights/last.pt' --name yolov4-csp-3-0.25 --resume

If you want to use multiple GPUs for training

python -m torch.distributed.launch --nproc_per_node 4 train.py --device 0,1,2,3 --batch-size 64 --data ./data/data.yaml --cfg yolov4-csp-3-0.25.cfg --weights yolov4-csp.weights --name yolov4-csp-3-0.25 --sync-bn

Counting Tracking and Creating submission csv file

Structure

Directory structure:

• Dataset_A (link)
• ScaledYOLOv4-yolov4-csp
• mish-cuda-master
• source_code
• weights

1. Go to 'AIC21-track1-team19/weigths'
2. Download weights: link
3. Go to 'AIC21-track1-team19/source_code/'
4. Run code python run_0304.py
5. Output file is in ./submission_output/submission.txt