tao-jiajun/mskpd-rdbca

MSKPD-RDBCA

MSKPD-RDBCA

Pre-trained models

You may download all models reported in the paper from Google Drive or Baidu Cloud.

We have trained our model with backbone of ResNet-18, ResNet-50 and MobileNet V2.

• Google (https://drive.google.com/drive/folders/1wmziQIaOQz1Pr_OndeWygZc3K81w7s-6?usp=sharing)

• Baidu (https://pan.baidu.com/s/12CDz25jR4X_6CYqcSd13yw the access code is：mskp)

Experimental environment

• Windows Server 2019
• Python 3.7
• Pytorch 1.8.1
• CUDNN 8.2.1.32

How to use

git clone https://github.com/Tao-JiaJun/MSKPD-RDBCA.git
cd mskpd-rdbca/

Please read the comment in eval_voc.py, and use the correct model and correct config files. For instance, if you want to evaluate the model with ResNet-18 at resolution of 384x384, please change the cfg file first.

# line 66
model = Detector(device, input_size=voc['input_size'], num_cls=20, strides = voc['strides'], scales=voc['scales'], cfg=RES18_RDB_384)

Second, select the model correspongding to the config file.

# line 70
checkpoint = torch.load('./weights/RES18_RDB_384.pth',map_location=device)

Third, run eval_voc.py

python eval_voc.py

Last, get the mean average precision.

python evel/get_map.py