Created by Junghyun Hong and Daijin Kim at POSTECH IM LAB
Detect and recognize traffic signs based on CNN.
This is based on Tsinghua Tencent 100k network (https://github.com/asyncbridge/tsinghua-tencent-100k).
We modified the network for korean traffic signs.
Becasue korean traffic sign datasets usually does not contain ground-truth for mask, we do not use pixel data.
Network is only composed of detection branch and classification branch.
You can download trained weight file from following link.
Mean file (https://drive.google.com/open?id=1CKp6UAAwGQoDHghs-URS2t6DcVBAr9Zm)
Trained weight file
- trained.caffemodel (https://drive.google.com/open?id=1o54zOxlQbfD_-JLYmOWTS8k_NxFxUfvh)
- trained.solverstate (https://drive.google.com/open?id=12D-MObATjNXT18OcRk8QQANsV82yWuBH)
Install caffe and pycaffe first.
Or use the pre-built library that is uploaded in this repository (python folder).
Refer demo.py how to run this program.