every blog every motto: You can do more than you think.
https://blog.csdn.net/weixin_39190382?type=blog
0. Preface
Extremely fast segmentation and labeling tool
1. Text
1.1 Installation
Create a virtual environment
conda create -n ISAT_with_segment_anything python==3.8
conda activate ISAT_with_segment_anything
install segment anything
git clone https://github.com/facebookresearch/segment-anything.git
cd segment-anything
pip install -e .
cd ..
Install ISAT_with_segment_anything
git clone https://github.com/yatengLG/ISAT_with_segment_anything.git
cd ISAT_with_segment_anything
pip install -r requirements.txt
Download the pre-trained model
https://github.com/yatengLG/ISAT_with_segment_anything
https://github.com/yatengLG/ISAT_with_segment_anything
1.2 run
python main.py
1.3 Operating instructions
1.3.1 Basic description
- Left button: select the area of interest
- Right click: deselect (area not of interest)
- E: end annotation, select category
- Drag and drop the vertices of the polygon to fine-tune the annotation.
- Adjust the occlusion relationship between targets through the height of the target layer (when there are overlapping areas between multiple targets)
1.3.2 Common shortcut keys
- q: Enable model annotation mode
- c: start normal annotation mode
- e: end annotation, select category
- s: save
- d: next picture
- a: previous picture
1.4 Precautions
- Only supports opening pictures under the file, does not support pictures in recursive subfolders
- If multiple folders are labeled in batches, you need to close the program and run it again. If the current folder is marked, do not close the program, open (load) a new file, if the picture names in the two folders are duplicated, confusion may occur
- For poorly labeled images, you need to use the regular labeling mode
reference
[1] [https://github.com/yatengLG/ISAT_with_segment_anything