Skip to main content

Quickly draw segmentation dataset

Project description

PyplotAnnotation

Introduction

The goal of the library is to create a software for quick annotation. More specifically, it is used for the generation of a segmentation dataset. The application shows images in a pyplot window and asks the user to choose a class, choose points to draw a polygon, and validate.

The application support up to nine classes, multiple polygons per label, and multiple labels per class.

Requirements

  • matplotlib
  • pillow
  • fire

How to use

The command line

To run the application, write the following command to the cmd:

python PyplotAnnotation {arg1} {arg2} {arg3}
  • {arg1} describes the path of the picture to segment, as recognized by glob. The default value is "*.png", which correspond to the .png in the current working directory.
  • {arg2} is the name of the folder in which the input of the dataset will be written.
  • {arg3} is the name of the folder in which the output of the dataset (the masks) will be written.

Example:

python -m PyplotAnnotation.PyplotAnnotation *.png inputs masks

Legal actions

The application follows a quite simple automaton:

New image state (initial state)

Actions performed on entering the state
  • A new picture is plotted.
  • The current polygons are initialized to a void list.

If coming from Polygon validated:

  • Save the picture inside the input_folder.
  • Save the masks (one per used class) to the mask_folder.

If these files already exist, they will be overwritten.

Leaving the state
Input State
1, 2, 3, 4, 5, 6, 7, 8, 9 Class selected

Class selected state

Actions performed on entering the state
  • The class of the current polygon is set to the input used to access this state.
Leaving the state
Input State
1, 2, 3, 4, 5, 6, 7, 8, 9 Class selected
Central click First point put

First point put state

Actions performed on entering the state

If coming from Class selected:

  • A point is added to the polygon.
  • The point is plotted on the picture.

If coming from Second point put:

  • The second point is removed.
  • The line between the second and first point is removed.
Leaving the state
Input State
Central click Second point put
Right click Class selected

Second point put state

Actions performed on entering the state

If coming from First point put:

  • A point is added to the polygon.
  • The point is plotted on the picture.
  • A line between the first and second point is plotted.

If coming from n-th point put:

  • The third point is removed.
  • The line between the third and second point is removed.
Leaving the state
Input State
Central click n-th point put
Right click First point put

n-th point put state

Actions performed on entering the state

If coming with a central click:

  • A point is added to the polygon.
  • The point is plotted on the picture.
  • A line between the two last point is plotted.
  • If coming from Second point put, set n at 3
  • If coming from n-th point put, set n at n+1

If coming with a right click:

  • The last point is removed.
  • The line between the last and second last point is removed.
  • Set n at n-1.
Leaving the state
Input State
0 Polygon validated
Central click n-th point put
Right click n-th point put
Right click Second point put

When using "Right click" : if n is three, go to "Second point put"-state, else go to "n-th point put".

Polygon validated state

Actions performed on entering the state

The current polygon is saved (now, it cannot be modified). A new polygon is initialized.

Leaving the state
Input State
Enter New image
1, 2, 3, 4, 5, 6, 7, 8, 9 Class selected

Format of the output

After the acceptation of each picture, multiple files are created.

  • First, a new image is generated in the input_folder, with the same name as the picture except that the extension is forced to be .png.
  • Second, the mask files are generated. Localised in the mask_folder, one of them as the same name as the same name as the input : this is the concatenation of each mask. Other files are created, one by class used. For each mask, the background is set at 0 while the mask is set at 255. Of course, the masks only have one canal each.

Examples

Credit for xiao_qiang (#pixiv1754781) for drawing the picture we use in example.

So, here is the figure plotted by the program. You can see the polygons created for the classes 1, 2 and 9. The 9-th class act as a eraser of any other layer. On the side, the legend contains the color of each class and the currently selectionned class.

Multiple polygons can be draw for a single label. You can use multiple labels on a single images, and also put points outside the picture.

In the mask_folder, the following files are created.

The first one is the concatenation of each class. The two other the segmentation of the classes 1 and 2. The class 9 has no file of his own but the mask of the first class is altered by the erasion of a part.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

PyplotAnnotation-1.1.1.tar.gz (18.4 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page