MediAug / Git / Diff of /README.md

Models:

DanielG/

MediAug

Downloads: 1

Diff of /README.md [05e710] .. [dd6877]

Switch to unified view


# MediAug

## Overview

MediAug is a set of tools for data augmentation of histology
slides. It is primaraly developed for cervical cancer by
augmenting Pap smear slides. However, it can be extended to
any cell data that has an image and mask of different types of
cells. Currently supports general image augmentation techniques
as well as specialized ones like cell insertion and blending.



## Installation

To install:

```bash
$ git clone https://github.com/smwade/MediAug
$ python setup.py install
```

## Datasets

There are two main open datasets for Pap smear images and MediAug is able to support both.

###  SMEAR

The SMEAR dataset is 917 indavidual cells. They are segmented by nucleus and cytoplasm.

<https://mde-lab.aegean.gr/downloads>

### SPIaKMeD

The SIPaKMeD Database consists of 4049 images of isolated cells that have been manually cropped from 966 cluster cell images of Pap smear slides. These images were acquired through a CCD camera adapted to an optical microscope. The cell images are divided into five categories containing normal, abnormal and benign cells.

<http://cs.uoi.gr/~marina/sipakmed.html>


## Custom Dataset

The data pipeline can work with other datasets besides SIPaKMed and SMEAR. In order to
use another, you must convert the data to the correct format.

```
slides/
  metaplastic/
    image/
    mask/
  parabasal/
    image/
    mask/
  ...
```

And for cells:

```
cells/
  metaplastic/
    image/
    mask/
  parabasal/
    image/
    mask/
  ...
```

## Notebooks

To show the library in action there are several notebooks that address key aspects of the library, such as what is a dataset, using Operations, and creating a Pipeline. These are found in `notebooks/`

## CLI

MediAug comes with a CLI with useful scripts. These include:

* generate-augment-dataset
* prepare-pix2pix-images
* resize-images

The list of all can be seen with the command

```bash
$ mediaug --help
```

### Generate cell augmented dataset

```bash
$ mediaug generate-augment-dataset --slide_dir <slide_dir> --cell_dir <cell_dir> --out_dir <out_dir> --num 1000 --max_cells <10>
```

### Prepare images for Pix2Pix

```bash
$ mediaug prepare-pix2pix-images --image_dir <image_dir> --mask_dir <mask_dir> --out_dir <out_dir> --split_ratio <split_ratio>
```

### Recursivly resize all images in directory

```bash
$ mediaug resize-images --input_dir <input_dir> --out_dir <out_dir> --w 256 --height 256
```

	a/README.md		b/README.md
1	# MediAug	1	# MediAug
2		2
3	## Overview	3	## Overview
4		4
5	MediAug is a set of tools for data augmentation of histology	5	MediAug is a set of tools for data augmentation of histology
6	slides. It is primaraly developed for cervical cancer by	6	slides. It is primaraly developed for cervical cancer by
7	augmenting Pap smear slides. However, it can be extended to	7	augmenting Pap smear slides. However, it can be extended to
8	any cell data that has an image and mask of different types of	8	any cell data that has an image and mask of different types of
9	cells. Currently supports general image augmentation techniques	9	cells. Currently supports general image augmentation techniques
10	as well as specialized ones like cell insertion and blending.	10	as well as specialized ones like cell insertion and blending.
11		11
12	![example_cell](docs/project_writeup/images/augment/example_cell.png)
13
14	## Installation	12	## Installation
15		13
16	To install:	14	To install:
17		15
18	```bash	16	```bash
19	$ git clone https://github.com/smwade/MediAug	17	$ git clone https://github.com/smwade/MediAug
20	$ python setup.py install	18	$ python setup.py install
21	```	19	```
22		20
23	## Datasets	21	## Datasets
24		22
25	There are two main open datasets for Pap smear images and MediAug is able to support both.	23	There are two main open datasets for Pap smear images and MediAug is able to support both.
26		24
27	### SMEAR	25	### SMEAR
28		26
29	The SMEAR dataset is 917 indavidual cells. They are segmented by nucleus and cytoplasm.	27	The SMEAR dataset is 917 indavidual cells. They are segmented by nucleus and cytoplasm.
30		28
31	<https://mde-lab.aegean.gr/downloads>	29	<https://mde-lab.aegean.gr/downloads>
32		30
33	### SPIaKMeD	31	### SPIaKMeD
34		32
35	The SIPaKMeD Database consists of 4049 images of isolated cells that have been manually cropped from 966 cluster cell images of Pap smear slides. These images were acquired through a CCD camera adapted to an optical microscope. The cell images are divided into five categories containing normal, abnormal and benign cells.	33	The SIPaKMeD Database consists of 4049 images of isolated cells that have been manually cropped from 966 cluster cell images of Pap smear slides. These images were acquired through a CCD camera adapted to an optical microscope. The cell images are divided into five categories containing normal, abnormal and benign cells.
36		34
37	<http://cs.uoi.gr/~marina/sipakmed.html>	35	<http://cs.uoi.gr/~marina/sipakmed.html>
38		36
39		37
40	## Custom Dataset	38	## Custom Dataset
41		39
42	The data pipeline can work with other datasets besides SIPaKMed and SMEAR. In order to	40	The data pipeline can work with other datasets besides SIPaKMed and SMEAR. In order to
43	use another, you must convert the data to the correct format.	41	use another, you must convert the data to the correct format.
44		42
45	```	43	```
46	slides/	44	slides/
47	metaplastic/	45	metaplastic/
48	image/	46	image/
49	mask/	47	mask/
50	parabasal/	48	parabasal/
51	image/	49	image/
52	mask/	50	mask/
53	...	51	...
54	```	52	```
55		53
56	And for cells:	54	And for cells:
57		55
58	```	56	```
59	cells/	57	cells/
60	metaplastic/	58	metaplastic/
61	image/	59	image/
62	mask/	60	mask/
63	parabasal/	61	parabasal/
64	image/	62	image/
65	mask/	63	mask/
66	...	64	...
67	```	65	```
68		66
69	## Notebooks	67	## Notebooks
70		68
71	To show the library in action there are several notebooks that address key aspects of the library, such as what is a dataset, using Operations, and creating a Pipeline. These are found in `notebooks/`	69	To show the library in action there are several notebooks that address key aspects of the library, such as what is a dataset, using Operations, and creating a Pipeline. These are found in `notebooks/`
72		70
73	## CLI	71	## CLI
74		72
75	MediAug comes with a CLI with useful scripts. These include:	73	MediAug comes with a CLI with useful scripts. These include:
76		74
77	* generate-augment-dataset	75	* generate-augment-dataset
78	* prepare-pix2pix-images	76	* prepare-pix2pix-images
79	* resize-images	77	* resize-images
80		78
81	The list of all can be seen with the command	79	The list of all can be seen with the command
82		80
83	```bash	81	```bash
84	$ mediaug --help	82	$ mediaug --help
85	```	83	```
86		84
87	### Generate cell augmented dataset	85	### Generate cell augmented dataset
88		86
89	```bash	87	```bash
90	$ mediaug generate-augment-dataset --slide_dir <slide_dir> --cell_dir <cell_dir> --out_dir <out_dir> --num 1000 --max_cells <10>	88	$ mediaug generate-augment-dataset --slide_dir <slide_dir> --cell_dir <cell_dir> --out_dir <out_dir> --num 1000 --max_cells <10>
91	```	89	```
92		90
93	### Prepare images for Pix2Pix	91	### Prepare images for Pix2Pix
94		92
95	```bash	93	```bash
96	$ mediaug prepare-pix2pix-images --image_dir <image_dir> --mask_dir <mask_dir> --out_dir <out_dir> --split_ratio <split_ratio>	94	$ mediaug prepare-pix2pix-images --image_dir <image_dir> --mask_dir <mask_dir> --out_dir <out_dir> --split_ratio <split_ratio>
97	```	95	```
98		96
99	### Recursivly resize all images in directory	97	### Recursivly resize all images in directory
100		98
101	```bash	99	```bash
102	$ mediaug resize-images --input_dir <input_dir> --out_dir <out_dir> --w 256 --height 256	100	$ mediaug resize-images --input_dir <input_dir> --out_dir <out_dir> --w 256 --height 256
103	```	101	```