pbaa 1.0.4

Prompt based automatic annotation

pbaa : Prompt-Based Automatic Annotation

Easy inference implementation of Grounded-SAM for Prompt-based automatic annotation

---

Table of Contents

• Installation
• Usage
• Gradio
• Demo
• License
• Acknowledgements

Installation

Docker (Recommend)

git clone https://github.com/dh031200/pbaa.git
docker build docker -t pbaa:latest
docker run --gpus all -it --ipc=host -v `pwd`:/workspace -p 7860:7860 pbaa:latest

Without docker

The code requires python>=3.8, CUDA==11.7.

pip install pbaa

Usage

Options

Usage: pbaa [OPTIONS]

Options:
  --version                    Show the version and exit.
  -s, --src TEXT               Source image or directory path
  -p, --prompt <TEXT TEXT>...  Space-separated a pair of prompt and target
                               classe. (Multi)
  -b, --box_threshold FLOAT    Threshold for Object Detection (default: 0.25)
  -n, --nms_threshold FLOAT    Threshold for NMS (default: 0.8)
  -o, --output_dir TEXT        Path to result data (default: 'outputs')
  -g, --gradio                 Launch gradio app
  -h, --help                   Show this message and exit.

CLI

# pbaa -s <Source> -p <prompt> <class> -p <prompt> <class> ...

pbaa -s source_image.jpg -p "black dog" dog
pbaa -s source_image.jpg -p "black dog" dog -p "white cat" cat

Python

from pbaa import PBAA

annotator = PBAA()
# inference(<Source path>, <prompt:class dict>, box_threshold=0.25, nms_threshold=0.8, save=None, output_dir="outputs")
annotator("path/to/source_image.jpg", {"black dog": "dog", "white cat": "cat"})

Gradio

Run the gradio demo with a simple command

pbaa -g

Output

Launch gradio app
Running on local URL:  http://0.0.0.0:7860

You can now access Gradio demos using your browser. localhost:7860

Demo

# Source : assets/demo9.jpg
# prompts : {"plant" : "plant", "picture" : "picture", "dog": "dog", "lamp" : "lamp", "carpet" : "carpet", "sofa" : "sofa"}

pbaa -s assets/demo9.jpg -p plant plant -p picture picture -p dog dog -p lamp lamp -p carpet carpet -p sofa sofa

Origin	Detection	Segmentation

Result data

demo9.json

json structure

filename
prompt
index
  ├ cls : class name
  ├ conf : confidence score
  ├ box : bounding box coordinates
  └ poly : polygon coordinates

License

pbaa is distributed under the terms of the Apache-2.0 license.

Acknowledgements

Grounded-Segment-Anything : https://github.com/IDEA-Research/Grounded-Segment-Anything
Grounding DINO : https://github.com/IDEA-Research/GroundingDINO
Segment-anything : https://github.com/facebookresearch/segment-anything