4995DeepLearning Competition Code

Technology:

Fine-tune: The pretrained model can be a good start point. Dehaze model shares the same low level feature. Freeze VGG layer for AECR-NET. Freeze no layers for YOLOv5.

Dataset Used:

AECR-NET: Dense-Haze, A2I2. YOLOv5: Vehicle Dataset( https://b2n.ir/vehicleDataset ), COCO, A2I2.

Dependencies

Basically this work is done in two virtual machines. Thus different dependencies is needed:

Dependencies 1

• Python3.7
• CUDA 10.2
• torch 1.4.0
• torchvision 0.5.0
• tqdm
• tensorflow
• pillow

Dependencies 2

• python3
• torch>=1.7.0
• torchvision>=0.8.1
• tqdm>=4.41.0

Dehaze AECR-NET

1: Install DCNv2:

cd DCNc2
bash ./make.sh

2: Install requirements

pip install -r requirement.txt

3: Train

python trainin.py

Referring for options if you want to change specific terms.

4: Test

python test.py 

YOLOv5

1: Install requirements

pip install -r requirement.txt

2: Install wandb

pip install wandb

3: Create dataset.yaml for A2I2 dataset

path: Dataset/dataset-vehicles  # dataset root dir
train: images/train  # train images (relative to 'path')
val: images/val  # val images (relative to 'path')
test:  # test images (optional)

# Classes
nc: 1  # number of classes
names: [ 'Vehicle']  # class names

4: Fine-tuning on a pre-trained model of yolov5.

python train.py --img 640 --batch 16 --epochs 50 --data dataset.yaml --weights yolov5m.pt

5: After train, gives you weights of train and you should use them for test.

python detect.py --weights runs/train/exp1/weights/best.pt --source test_images/