# PathologyGo

Core components of PathologyGo, the AI assistance system designed for histopathological inference.

<b>Dependency</b>

* Docker

* Python 2.7 and 3.x

* openslide

* tensorflow_serving

* grpc

* pillow

* numpy

* opencv-python

<b>Dockerized TensorFlow Serving</b>

* GPU version: [GitHub](https://github.com/physicso/tensorflow_serving_gpu), [Docker Hub](https://hub.docker.com/r/physicso/tf_serving_gpu).

* CPU version: [Docker Hub](https://hub.docker.com/r/tensorflow/serving).

<b>Quick Start</b>

This code is easy to implement. Just change the path to your data repo:

```python

from utils import config

GPU_LIST = config.INFERENCE_GPUS

import os

os.environ["CUDA_VISIBLE_DEVICES"] = ','.join('{0}'.format(n) for n in GPU_LIST)

from inference import Inference

if __name__ == '__main__':

    pg = Inference(data_dir='/path/to/data/', data_list='/path/to/list',

                    class_num=2, result_dir='./result', use_level=1)

    pg.run()

```

You may configure all the model-specific parameters in `utils/config.py`.

<b>Example</b>

Use the [CAMELYON16](https://camelyon16.grand-challenge.org/) test dataset as an example, 

the data path should be `/data/CAMELYON/`, and the content of the data list is

```

001.tif

002.tif

...

```

The predicted heatmaps will be written to `./result`.

<b>DIY Notes</b>

You may use other exported models. You can change the model name for TensorFlow Serving in `utils/config.py`. Just remember to modify `class_num` and `use_level`.

Note that the default input / output tensor name should be `input` / `output`.