AutoLabel

This project is developed using PyQt and is inspired by SAM, creating an automatic annotation system based on the concept of LabelImg.

Features

Similar functionalities to LabelImg (although very basic).

Annotate the rect box

Automated annotations with SAM

1. Right mouse on the object you're interested in.

2. Left mouse to exclude the objects you're not interested in.

Saving

You can choose from three formats: YOLO, Pascal VOC and COCO, the default is YOLO.
Notice:

• If you choose the COCO format, it must be consistent from the beginning, otherwise, you will only obtain annotations for a single image.

Visualization

You can place the annotations in the same directory as the images, and the labels file name must be same with image file name.
If it is in COCO format, put annotations.json in the directory.

Hotkeys

Hotkey	Description
D	Next image
A	Previous image
Space	Automatically predict rectangle box
del	Delete the selected rectangle box
Ctrl + S	Save annotations
↑→↓←	Move the selected rectangle box

Environment

• PyQt5
• lxml
• Segment Anything

Model

Name	Checkpoint
vit_h	ViT-H SAM model
vit_l	ViT-L SAM model
vit_b	ViT-B SAM model

Then modify the name and path here:

self.sam = SA(model_name="vit_b", model_path="sam_vit_b_01ec64.pth")

Usage

1. Your directory must include classes.txt and you can edit the classes on your own.

   dog
   person
   cat
   ...

2. Run

   python autoLabel.py

Reference

• segment-anything
• labelImg