Model - VGG-16 Nail part detection model for faster-RCNNs
Model - Hand and Foot Image Selector
Model - Fine Image Selector

- Linux (Ubuntu)
- NVIDIA GPU (GTX-1050 or better)
- BVLC PyCaffe
- py-faster RCNNs

- VGG-16 nail part detection model : #1 FigShare (483 Mbytes)
- demo_nail.py : FigShare

1. It is difficult to compile a CPU-mode faster-rcnn on Windows operating system at present.
NVidia GPU with CUDA and cuDNN is required because it takes too much time to conduct CNNs training without GPU
We recommend to install py-faster-rcnn program (https://github.com/rbgirshick/py-faster-rcnn) which operated on Linux (http://ubuntu.com).
Installation Tutorial by Huangying : https://huangying-zhan.github.io/2016/09/22/detection-faster-rcnn.html

2. Download (VGG-16 nail part detection model) : #1 FigShare (483 Mbytes)
Model - VGG-16 nail part detection ; 2 outputs(class) : #0 background #1 nail

The VGG-16 nail part model was trained using information about the crop location on the nail part from the Asan A2 dataset as instructed by the following tutorials.
http://sgsai.blogspot.kr/2016/02/training-faster-r-cnn-on-custom-dataset.html
https://github.com/deboc/py-faster-rcnn/tree/master/help

3. We modified demo.py of py-faster-rcnn (https://github.com/rbgirshick/py-faster-rcnn/blob/master/tools/demo.py) to get the following image.

Download our demo_nail.py ( FigShare )





- BVLC PyCaffe for Python example

- Caffe Model + Python source + WIN64 EXE example : #1 FigShare (380 Mbytes)

ResNet-152; 3 outputs : #0 handfoot photos #1 others #2 index photos

We created a Windows 64bit executable using pyinstaller (http://www.pyinstaller.org).

Example)

1) Windows 64bit EXE example
c:\handfoot\demo.bat

2) Python example ; requirements - PyCaffe
handfoot.bat [source_path] [result_prefix]


- BVLC PyCaffe

- Model Fine image selector : #1 FigShare (207 Mbytes)

ResNet-152; 6 outputs
#0 nail
#1 non-focused or tilted nail
#2 skin photo, not nail
#3 skin photo, not nail
#4 index photo, or EMR screenshot
#5 non-skin photo, general internet images