lvis-api - Refined API and Dataset for Large Vocabulary Instance Segmentation

LVIS API: Large Vocabulary Instance Segmentation Dataset

LVIS, pronounced 'el-vis', is a pioneering dataset designed for Large Vocabulary Instance Segmentation. It aims to provide comprehensive, high-quality instance segmentation masks across a broad array of object categories. Once fully developed, LVIS will house over 2 million instance segmentation masks spanning more than 1,200 object categories, collected from 164,000 images. The LVIS API serves as a powerful tool for interacting with these rich annotation files, enabling users to read, visualize, and evaluate segmentation results efficiently.

Overview of LVIS v1.0

The first major release, LVIS v1.0, includes annotations for 159,623 images, divided into training (100,000 images), validation (20,000 images), test-development (20,000 images), and test-challenge (20,000 images) sets. This release is publicly accessible through the LVIS website and plays a central role in the second LVIS Challenge, presented at the Joint COCO and LVIS Workshop during ECCV 2020.

Setting Up the LVIS API

Setting up the LVIS API is straightforward. Users can establish a virtual environment and install the necessary components via pip. Here’s a quick guide:

Create and Activate a Virtual Environment:

python3 -m venv env               # Create a virtual environment
source env/bin/activate           # Activate the virtual environment

Install Required APIs: Before installing the LVIS API, the COCO API must be installed, which requires numpy. Ensure numpy is installed:
```
pip install 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'
pip install lvis
```
Deactivate the Environment After Use:
```
deactivate  # Exit virtual environment
```

For users who prefer to clone the repository, the installation within the repo follows similar steps:

python3 -m venv env               # Create a virtual environment
source env/bin/activate           # Activate the virtual environment

pip install 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'
pip install .
python test.py                    # Test if the installation was correct
deactivate  # Exit virtual environment

Research and Citing LVIS

The LVIS dataset and API are essential resources for researchers focusing on computer vision and instance segmentation. If you utilize LVIS in your own research, it is encouraged to cite the following paper to acknowledge the creators:

@inproceedings{gupta2019lvis,
  title={{LVIS}: A Dataset for Large Vocabulary Instance Segmentation},
  author={Gupta, Agrim and Dollar, Piotr and Girshick, Ross},
  booktitle={Proceedings of the {IEEE} Conference on Computer Vision and Pattern Recognition},
  year={2019}
}

Contributions and Development

The LVIS API is built upon the foundational work of the COCO PythonAPI. While it shares core functionalities with COCO, it incorporates specific changes to support LVIS and its extensive dataset.

LVIS represents a significant step forward in the field of computer vision, providing detailed, high-quality data and tools for researchers working with vast and varied object categories in instance segmentation.