deeva 1.0.1

Object Detection Data Analysis Toolbox

Deeva 🚀

Your Smart Analytics Companion for Object Detection Datasets

🎯 Overview

Deeva is a powerful yet easy-to-use analytics toolkit that makes exploring Object Detection datasets a breeze, whether you're just starting out or a seasoned pro.

Built with Streamlit, it offers an intuitive interface packed with features that let you dive into your data quickly or take a deeper look when you need it. Deeva is designed to simplify data exploration and reporting, so you can get meaningful insights without the hassle.

Key Features

• 💻 Run locally: Launch effortlessly on your local machine for seamless, offline use.
• 🚀 Instant Setup: Quickly start visualizing data by pointing Deeva to a specific dataset folder.
• 📊 Rich Interactive Dashboards: Build insightful, interactive dashboards for rich data exploration with minimal effort.
• 🎨 Customizable CLI: Use simple command-line commands to launch Deeva with flexible paths and configurations.
• 💾 Smart Caching: Efficient processing with intelligent data caching for large datasets
• 🎲 Built-in Toy Datasets: Quickly get started with the included coco128 dataset, perfect for initial experimentation.

🛠 Installation

install with pip:

$ pip install deeva

Alternatively, use a virtual environment (recommended):

$ python3 -m venv myenv
$ source myenv/bin/activate

$ pip install deeva

⚡ Quickstart

After installation, launch Deeva by running:

$ deeva start

This will open the input page where you can specify the data path.

Data structure

Your dataset folder should look like this:

data-path/
├── images/        # Folder containing image files (e.g., .jpg, .png)
├── labels/        # Folder containing label files (e.g., .txt, .xml)
└── labelmap.txt   # A file mapping class IDs to class labels (optional)

💡 Insights & Analytics

Deeva offers a powerful set of statistical insights to give you a detailed understanding of your dataset, including:

1. File Matching and Integrity

• Image-Label Matching: Calculates how many images have corresponding labels (and vice versa).
• Filename Consistency: Identifies misaligned or corrupted files in images and labels.
• Data Cleaning: Provides tools to identify and isolate mismatched or corrupted files.

2. Dataset Overview

• File Formats & Backgrounds: View format distribution (yolo vs. voc, jpeg vs. png).
• Class Distribution: Displays instance counts and images per class, highlighting any class imbalances.
• Class Co-occurrence: Shows how frequently different classes appear together.

3. Annotation Insights

• Bounding Box Analysis: Provides insights into box center, width/height, and median box sizes.
• Box Size Distribution: Analyzes box size categories with adjustable thresholds for small, medium, and large sizes.

4. Image Statistics

• Color Analysis: Displays dominant colors and their tones extracted from images.
• Image Dimensions: Examines height, width, and aspect ratios across your dataset.
• CBS (Contrast, Brightness, Saturation): Shows contrast, brightness, and saturation distributions across the dataset.

5. Overlap Statistics

• Cases: Classify and cluster overlapping instances from two specific classes into n predefined cases. Display representative example images for each case to help visualize typical overlap patterns.
• Ratios: Calculate and visualize the overlap ratio distributions for each class
• With/without overlaps: Present a side-by-side comparison of images and co-occurrences with and without overlaps

🔖 Caching & Version control

Deeva employs efficient caching to streamline your data processing workflow. For large datasets, users have the option to sample a subset of the data—allowing for quicker initial exploration.

Data extracted during time-consuming operations can be saved as a dataframe on disk for effortless access in future sessions, enabling a faster, more efficient experience by skipping redundant processing steps.

To track different versions of your dataset you need to simply put them into different folders and Deeva will do the rest

🌟 Contributing

Deeva welcomes contributions! If you have ideas or want to add new features, please feel free to open a pull request or start a discussion on GitHub.

License

Deeva is completely free and open-source and licensed under the Apache 2.0 license.