Chitra is an open-source, multi-functional library for full-stack deep learning that simplifies model building, API development, and model deployment. Published on GitHub under the Apache-2.0 license with about 230 stars and updated in October 2026, it is installed from PyPI and documented on Read the Docs.
Chitra tries to cover the whole life of an image deep learning project in one library. A Chitra image class loads pictures from a URL, file path, or NumPy array and plots bounding boxes with little code. The trainer module enables rapid experiments with different models, with faster data loading, progressive resizing of images, and model interpretation through GradCAM and GradCAM++ without extra code. When a model is ready, Chitra can create an interactive UI prototype or a REST API backend for serving it in production in a framework-agnostic way, and it can automatically dockerize models for deployment. The project positions itself as a single toolkit spanning data visualization, training, explainability, and serving.
Installation is flexible: a minimal pip install covers the basics, while optional extras add the training stack or the serving stack depending on which parts of the library a project actually needs. The documentation walks through the image class, the trainer module, model interpretation, and the model server step by step, which makes it possible to adopt Chitra one component at a time rather than all at once.
Chitra is free and open source under the Apache-2.0 license. Install it with pip in minimal, full, training, or serving configurations; there is no paid tier or hosted service.
Chitra is for Python developers and data scientists, especially in computer vision, who want one library that takes a model from experiment to an explainable, served API.