$ show PINTO0309/PINTO_model_zoo
This repository is a large model zoo (500+ models) focused on converting and quantizing pre-trained computer vision models for edge deployment across multiple frameworks (TFLite, ONNX, OpenVINO, CoreML, TensorRT). It provides hundreds of download scripts for pre-converted model files, conversion pipelines (PyTorch/ TensorFlow to ONNX/TFLite), and a modest set of inference demo scripts for real-time video processing. The repo is primarily a research/engineering resource for edge AI practitioners, not a software application. It exhibits heavy duplication across model directories, relies on an external S3 bucket for model distribution, and uses older framework versions (TensorFlow 1.x/2.x, ONNX opset 11-13). Despite these maintenance concerns, the systematic conversion workflow and post-processing ONNX subgraph generators (e.g., NMS, box decoding) represent reusable patterns for anyone building end-to-end inference pipelines. The repo is hybrid in nature: while classified as data_research due to its focus on pre-trained model artifacts and lack of original training code, it also contains application-like demo code and CLI entry points.
The LLM identified 3 files worth highlighting for reuse:
A PyTorch-based ONNX export script that generates a standalone ONNX subgraph for grid decoding (anchor points and strides) used in YOLOX-style detectors. This component is isolated, parameterized by stride and grid size, and can be extracted and reused for any detection model requiring grid-based decoding. It demonstrates non-obvious skill in ONNX graph manipulation and post-processing pipeline construction.
A script that creates an ONNX subgraph to split detection outputs into boxes, scores, and optionally class scores using gather_nd operations. It is modular, configurable via command-line arguments (batch size, number of boxes/classes), and can be applied to any object detector that outputs a flat tensor. The pattern of slicing and gathering indices in ONNX is reusable for many post-processing tasks.
A script that generates ONNX operations to produce batch, class, and box index tensors needed for NMS output merging. It is a standalone component that solves a common problem in multi-batch object detection post-processing: aligning indices from NMS into a final output format. The pattern is reusable across any model that uses NMS and needs to output batch/class/box information.