gyasmeen-ml / spbl_pytorch

PyTorch implementation of "Self-Paced Balance Learning for Clinical Skin Disease Recognition"

SPBL_Pytorch

PyTorch implementation of "Self-Paced Balance Learning for Clinical Skin Disease Recognition"

Installation

This project is based on spaco and Open-Reid. And we add imbalanced learning methods in it and modify some code for compatibility.

Requirements

• Python 3.4+
• PyTorch 0.4.0

Setup the Environment and Packages

i. Create a new environment

We recommend Anaconda as the package & environment manager. And here is an example:

conda create -n spbl
conda activate spbl

ii. Install PyTorch

Follow the official instructions to install Pytorch. Here is an example using conda:

conda install pytorch=0.4.0 torchvision -c pytorch

iii. Install Cython

conda install cython 
# or "pip install cython"

iv. Install Python Matlab Engine

Follow the official instructions to install python matlab engine. Here is an example using conda:

cd "matlabroot/extern/engines/python"
python setup.py install

Install SPBL

i.Clone the repository

git clone https://github.com/xpwu95/SPBL_Pytorch.git

ii. Compile extensions

cd open-reid
python setup.py install

Prepare Data

The SD-198 and SD-260 datasets can be downloaded from http://xiaopingwu.cn/assets/projects/sd-198/.

It is recommended to symlink the datasets root to spbl/data/dataset_name/raw/images.

ln -s $YOUR_DATA_ROOT data/dataset_name/raw/images

The directories should be arranged like this:

SPBL_Pytorch
├──	spbl
|	├── reid
|	├── examples
|	│   ├── data
|	│   │   ├── sd-198
|	│   │   │   ├── raw
|	│   │   │   │   ├── images
|	│   │   │   │   ├── train.txt
|	│   │   │   │   ├── val.txt
|	│   │   ├── sd-260
|	│   │   ├── mit67
|	│   │   ├── caltech101
|	│   │   ├── minist
|	│   │   ├── mlc

Running

python spbl.py

Configuration

The configuration parameters are mainly in the /examples/cfg.py files. The parameters you most probably change are as follows:

• input_size: the size of input image
• lr: learning rate
• batch_size: batch size
• workers:
• iter_step: the pace parameter of the SPBL
• gamma:
• train_ratio: the ratio of the initial training set
• model: model name
• dataset: dataset name
• class_num: the class number of the dataset
• epochs: total training epochs
• step_size: the step size of the learning rate decay
• add_ratios: add ratios

Results

Dataset	Precision	Recall	F1	G-mean	MAUC	Accuracy
SD-198	71.4±1.7	65.7±1.6	66.2±1.6	42.8±4.0	68.5±1.6	67.8±1.8
SD-260	59.9±1.6	48.2±1.1	51.0±0.9	19.6±1.1	64.8±1.2	65.1±0.8

Dataset	MIT-67	Caltech-101	MINIST	MLC
Accuracy	64.1±0.5	88.6±0.4	99.0±0.1	72.0

License

This project is released under the MIT license.