Train a custom image segmentation or object detection model using Apple's production-grade research framework.
Download pre-trained OpenELM language model weights and run inference on an M-series Mac using MLX.
Reproduce results from Apple research papers using the included training configurations and Jupyter tutorials.
Fine-tune a vision-language model on a custom dataset using CoreNet's modular training architecture.
| apple/corenet | infrasys-ai/aiinfra | threestudio-project/threestudio | |
|---|---|---|---|
| Stars | 7,003 | 7,000 | 7,016 |
| Language | Jupyter Notebook | Jupyter Notebook | Jupyter Notebook |
| Setup difficulty | hard | easy | hard |
| Complexity | 4/5 | 3/5 | 5/5 |
| Audience | researcher | researcher | researcher |
Figures from each repo's GitHub metadata at analysis time.
Requires Git LFS installed before cloning, Python 3.10+ on Linux, and a compatible GPU for most training tasks.
CoreNet is a Python library from Apple for training AI models, specifically the kind called deep neural networks. It is the internal toolkit Apple researchers use to build and experiment with models that can recognize objects in images, segment images into regions, detect objects, and work with large language models. Several published Apple research papers were produced using this library, and the repository includes the training configurations needed to reproduce those results. The library supports a wide range of model types, from lightweight models designed to run efficiently on phones to larger foundation models used in vision-language tasks. Pre-trained model weights for many of these are available alongside the code. Tutorials and Jupyter notebooks are included to help newcomers get started, covering tasks like training a new model from scratch and running semantic segmentation or object detection. One notable included project is OpenELM, a family of efficient language models trained and released with open weights. Another is CatLIP, which achieves competitive image-text recognition at faster training speeds. These are packaged as separate project directories within the repository, each with documentation and configuration files. For developers who use Apple Silicon hardware, the repository also includes examples that run CoreNet models via MLX, Apple's framework optimized for Mac chips. This makes it possible to run inference locally on a Mac without needing a data-center GPU. Installation requires Python 3.10 or higher on Linux, or Python 3.9 or higher on macOS, along with a recent version of the PyTorch library. The repository uses Git LFS to store large binary files like datasets and checkpoints, so that tool must be installed before cloning. The library is oriented toward researchers and engineers who want to train or fine-tune models rather than end users looking for a finished product.
CoreNet is Apple's open-source Python library for training deep learning models, used internally for published research, covering image classification, segmentation, object detection, and large language models including OpenELM.
Mainly Jupyter Notebook. The stack also includes Python, PyTorch, Jupyter Notebook.
Setup difficulty is rated hard, with roughly 1day+ to a first successful run.
Mainly researcher.
This repo across BitVibe Labs
Verify against the repo before relying on details.