ayolo 0.2.1

PyQt5 based annotation tool for yolov4 datasets, providing fast and easy ways of annotating.

Ayolo

PyQt5 based annotation tool for yolov4 datasets, providing fast and easy ways of annotating.

Wait what, reinventing the wheel ? Well ... technically yes, but if it has improvements or features that others don't, then why not ?

If you like the tool, a star to this repository would be nice.

Bugs

For bugs (e.g. RuntimeError: cuda runtime error (59) : device-side assert triggered) you can go over to this repo's issue tracker, specially on the closed ones, to see if it's been resolved or its workaround. If you are sure that the bug is not reported, feel free to open a new issue for that.

Installation

• Requires Python 3.7 (or whatever is supported by PyQt5)
• No executables, build it from source if you need one.
• For global access, install it outside of virtual environments.

pip install ayolo

Usage

• Call annotate with the dataset path, either relative or absolute.

ayolo annotate <dir_path>

Dataset Structure

<dir_path>
├── img_1.png
├── img_2.png
├── img_3.png
├── classes.txt
└── annotations.txt

• Supported image extensions: .png, .jpg
• Supported annotations format: .txt
• Supported classnames format: .txt

Interfaces

This module provides a PyQt5 window for making annotations on yolov4 datasets. The annotator window consists of 3 interfaces:

Image Browser (Left)

• Filters for annotated and unannotated images.
• Keyboard and mouse shortcuts for faster navigation.
• Annotations are saved every time an image it's navigated away from.

Annotator (Center)

• Twice left click for annotating to avoid wrist damage.
• Cross lines for better accurary.
• Realtime coordinate location.
• Window resize persistent.
• Image scaled to fit container.
• Instant colors for class distinction.
• Right click to undo selection.

Control Panel (Right)

• Realtime current annotations listing.
• Buttons with global shotcuts for managing annotations/images.
• Search feature for class names.
• Ability to create new classes on the go.
• Keyboard and mouse shortcuts for navigating class names search results.

Buttons

Available buttons in annotator window:

Button	Action
Prev	Navigate previous Image
Next	Navigate next Image
Save	Save current annotations
Null	Mark image as Null
Undo	Undo last annotation
Delete	Delete/Discard current Image
Clear	Clear all current annotations

Keyboard Shortcuts

Available keyboard shortcuts in annotator window / interfaces:

Key/KeySequence	Interface	Action
PgUp	Global	Navigate previous Image
PgDn	Global	Navigate next Image
Ctrl + S	Global	Save current annotations
Ctrl + N	Global	Mark image as Null
Ctrl + Z	Global	Undo last annotation
Ctrl + D	Global	Delete/Discard current Image
Ctrl + X	Global	Clear all current annotations
ArrowUp	Control Panel	Select previous class
ArrowDn	Control Panel	Select next class

Mouse Shortcuts

Available mouse events in annotator interface:

Key/KeySequence	Interface	Action
Left	Annotator	Start/Finish rectangle annotation
Right	Annotator	Undo last annotation
ScrollUp	Annotator	Select previous class
ScrollDn	Annotator	Select next class
Back	Annotator	Navigate previous Image
Forward	Annotator	Navigate next Image

Back and Forward buttons are not available in some mice, they are typically only available on gaming mice

Contribute

Contributions are welcomed, as long as it doesn't cause any harm to users.

Possibly implemented features ?

I am a normal hooman who has a daily job, thus not a lot time available, but ... let's put some expectations ... soon :tm:.

• Augmentation of datasets: brightness, rotations, contrast, etc.
• Preprocessing of datasets: cropping, resizing, etc.
• Choose output format: yolov4 txt, coco, xml, etc.