PINTO_model_zoo is a model repository that stores neural network models converted between multiple deep learning frameworks.
The repository addresses the challenge of working with models across different frameworks by providing pre-converted versions of popular models. It supports conversion between TensorFlow, PyTorch, ONNX, OpenVINO, TensorFlow.js, TF-TRT, TensorFlow Lite (in Float32, Float16, and INT8 variants), EdgeTPU, CoreML, and other formats. The approach involves quantization work and framework interconversion to make models accessible in the format needed for a given deployment target.
This tool suits developers who need models in multiple framework formats without performing conversions themselves. It is particularly valuable for edge deployment scenarios where TensorFlow Lite, EdgeTPU, or CoreML formats are required. The repository emphasizes that each model folder contains its own license file, and users must review these before use, as the conversion scripts are MIT-licensed but source models retain their original licenses.
The project maintains an active collection of converted models across numerous computer vision architectures and frameworks. The maintainer has documented quantization techniques and conversion methodologies through detailed articles and has welcomed community contributions for sample code and additional conversions.
