Installation

The basic requirements for installing PixelSSL are:

NOTE: PixelSSL requires Nvidia GPU to run, i.e., the CPU only mode is currently not supported.

We recommand creating a new conda virtual environment for PixelSSL as follow:

Create a conda virtual environment named PixelSSL and activate it:
```
conda create -n PixelSSL python=3.6
source activate PixelSSL
```
Install PyTorch (>=1.0.0) and the corresponding torchvision following the PyTorch official instructions.
For example, if you use CUDA 8.0:
```
conda install pytorch==1.0.0 torchvision==0.2.1 cuda80 -c pytorch
```

Clone the repository of PixelSSL:

git clone https://github.com/ZHKKKe/PixelSSL.git
cd PixelSSL

Install other dependencies and PixelSSL:

We provide two options for using PixelSSL as follow:

(a) If you want to develop and validate a new semi-supervised learning algorithm (or try a new vision task) based on the latest code of PixelSSL, you need to install python dependencies:
```
pip install -r pixelssl/requirements.txt
```
(b) If you want to use the semi-supervised learning algorithms provided by PixelSSL in your own task project (should follow the task template in PixelSSL), you can compile and install PixelSSL into the current conda virtual environment:
```
pip install .
```
or
```
python setup.py install
```
Then, in any directory, you can import the package of PixelSSL in the current conda virtual environment:
```
python
>>> import pixelssl
```
After completing any of the above options, you can follow Getting Started to run the integrated task code.

Provide feedback