DataShotExtractor

An AI-based web application capable of extracting data from an image containing one or more tables and exporting it to various standard data formats (CSV, XML, JSON, etc.) or Markdown. The system will be able to queue different data extraction jobs, since the inference process is time-consuming. It will also be able to automatically detect, if available, the option to perform inference on a GPU; otherwise, it will use the default option of performing inference on a CPU.

The “Data Shot Extractor” project aims to develop a web application that, using artificial intelligence (AI), enables the accurate extraction of tabular data from images. This tool will be capable of processing images containing tables, segmenting them, and automatically extracting the information to then export it in various standard formats such as CSV, XML, JSON, or Markdown, facilitating its integration into different workflows.

The system will be designed to efficiently handle the computational load associated with AI inference, particularly in tasks involving segmentation and the recognition of complex patterns in images. To achieve this, a job queue will be implemented, allowing for the orderly management of multiple data extraction requests, thereby preventing system overload and optimizing processing time.

A key feature of the project will be the automatic detection of GPU availability for inference. If a GPU is detected, the system will use it to accelerate the process; otherwise, it will fall back to the CPU as the default option, ensuring that the application performs optimally in different environments.

El desarrollo de esta aplicación se llevará a cabo utilizando Python como lenguaje principal, PyTorch para el modelado y entrenamiento de los modelos de IA, y frameworks web para la creación de la interfaz de usuario. Además, se explorarán modelos preentrenados disponibles en plataformas como HuggingFace y se optimizarán para su despliegue utilizando ONNX, asegurando así un rendimiento eficiente en diferentes entornos.

“Data Shot Extractor” se posiciona como una solución versátil y práctica para la extracción automatizada de datos desde imágenes, dirigida a usuarios que necesitan integrar datos tabulares de forma rápida y precisa en sus sistemas, con la flexibilidad de adaptarse a diferentes capacidades de hardware.

 


TOP