VectorInstitute / vector_cv_tools

Tooling, demos, and examples from Vector Computer Vision Project

vector_cv_tools

Vector Computer Vision Project Tooling support

Introduction

This is a tool-kit provided by the AI Engineering team for the Computer Vision project at Vector Institute. It includes various datasets readily loadable from the shared cluster as well as useful image/video tools such as data augmentation and visualization utilities.

datasets

Provides a list of dataset used for Object Detection, Image Segmentation, and Video Recognition tasks.

Image:

• MSCOCO 2017: image captioning, detection, and segmentation
• Cityscape: segmentation
• MVTec: Anomoly detection and segmentation for common objects

Video:

• ActivityNet: Videos of human activities
• Kinetics-700: Videos including human-object interactions as well as human-human interactions.

transforms

• Various data augmentation transforms considered useful for CV tasks

Requirements

To install the requirements for the package, run

pip install -r requirements.txt
pip install pycocotools

Installation

To install the package, run

git clone https://github.com/VectorInstitute/vector_cv_tools.git
cd vector_cv_tools
pip install -e .

The team

This repository is primarily developed by Xin Li and Gerald Shen from Vector Institute, with contributions from Sheen Thusoo for the demo software.