cvpal 0.0.3

A Python library for every Computer Vision Engineer

CvPal

Introduction

Welcome to CvPal!👋

CvPal is a powerful Python package designed to assist Computer Vision engineers in performing a variety of image operations. The primary goal of cvPal is to simplify the process of image manipulation and dataset management, enabling engineers to focus on building and optimizing their machine learning pipelines.

With CvPal, you can easily handle tasks such as merging datasets, removing specific labels, counting label occurrences, and generating comprehensive reports on your dataset, all with a user-friendly interface and comprehensive documentation.

Features

• Dataset Merging: Functions to seamlessly merge different datasets, each with multiple attributes.
• Label Removal: Functions to remove specific labels from your dataset.
• Label Occurrence Counting: Functions to count the occurrences of specific labels.
• Dataset Reporting: Functions to generate comprehensive reports on your dataset.
• Easy Integration: Seamlessly integrate with existing ML pipelines.
• Comprehensive Documentation: Detailed documentation to guide you through all features and functionalities.

Dataset Folder Structure

To ensure CvPal works properly, the provided dataset folder should follow this structure:

• folder/
  • train/
    • images/
    • labels/
  • test/
    • images/
    • labels/
  • valid/
    • images/
    • labels/

Additionally, for datasets using TXT format, a data.yaml config file is required:

• folder/
  • data.yaml

Example data.yaml:

names:
  - Old_Paper
  - Rock
  - Scissors
nc: 3
roboflow:
  license: Private
  project: rock-paper-scissors-sxsw
  url: https://universe.roboflow.com/roboflow-58fyf/rock-paper-scissors-sxsw/dataset/14
  version: 14
  workspace: roboflow-58fyf
test: ../test/images
train: Rock-Paper-Scissors-SXSW-14/train/images
val: Rock-Paper-Scissors-SXSW-14/valid/images

Installation

You can install CvPal via pip:

pip install cvpal

Example Usage

from cvpal import CvPal

cp = CvPal()

# Read data from the specified directory
cp.read_data("/content/Rock-Paper-Scissors-SXSW-14", data_type="txt")

# Generate a comprehensive report on the dataset
cp.report()

Supported Models and Formats

Model Name	Supported Format	Support in Package
YOLOv3	Darknet TXT	❌
YOLOv4	Darknet TXT	❌
YOLOv5	TXT & YAML config	✅
YOLOv7	TXT & YAML config	✅
YOLOv8	TXT & YAML config	✅
YOLOv9	TXT & YAML config	✅
EfficientDet	Pytorch JSON annotations	❌
Detectron 2	JSON annotations	❌
Segmentation Models	XML format	❌
Tensorflow 1.5 Object Detection models	Binary format	❌
Tensorflow 2.0 Object Detection models	Binary format	❌
Fine-tune PaliGemma	JSONL annotations	❌
Apple's CreateML	Proprietary JSON format	❌
Turi Create tools	Proprietary JSON format	❌

Call for Contributions

We Need Your Help The cvPal project values your skills and passion!

We appreciate any small enhancements or fixes you can make. If you want to make bigger changes to the source code, please let us know through the mailing list first.

There are many other ways to contribute to zein besides writing code. You can also:

• Help us manage new and old issues
• Create tutorials, presentations, and other learning materials
• Evaluate pull requests