14.8K ·GitHub
What is this?
A collection of pretrained machine learning models ported to TensorFlow.js, covering vision, audio, text, and general utility tasks, ready to use directly in browser or Node.js projects.
A collection of pretrained machine learning models ported to TensorFlow.js, covering vision, audio, text, and general utility tasks, ready to use directly in browser or Node.js projects.Is this a single library or a set of separate packages?It is a collection of separate npm packages, one per model (e.g. mobilenet, pose-detection, coco-ssd), each independently installable.
It is a collection of separate npm packages, one per model (e.g. mobilenet, pose-detection, coco-ssd), each independently installable.How do I add a model to my project?Install the specific model package from npm, for example 'npm i @tensorflow-models/coco-ssd' or 'npm i @tensorflow-models/pose-detection', then import and call its API.
Install the specific model package from npm, for example 'npm i @tensorflow-models/coco-ssd' or 'npm i @tensorflow-models/pose-detection', then import and call its API.Do I need to know TensorFlow internals to use these models?No, the APIs are designed to hide tensors so non-machine-learning experts can use them directly.
No, the APIs are designed to hide tensors so non-machine-learning experts can use them directly.Can I use these models beyond out-of-the-box inference?Yes, most models also support transfer learning, and there's a KNN Classifier utility specifically for building custom classifiers on top of model outputs.
Yes, most models also support transfer learning, and there's a KNN Classifier utility specifically for building custom classifiers on top of model outputs.Where do these models run?They run client-side in the browser via TensorFlow.js, since they're distributed through NPM and unpkg for direct web use.
They run client-side in the browser via TensorFlow.js, since they're distributed through NPM and unpkg for direct web use.What kinds of models are included?Vision models (MobileNet, pose/hand/face landmark detection, Coco SSD, DeepLab v3, portrait depth estimation), an audio speech command classifier, text models (Universal Sentence Encoder, toxicity scoring), and a general-purpose KNN classifier utility.
Vision models (MobileNet, pose/hand/face landmark detection, Coco SSD, DeepLab v3, portrait depth estimation), an audio speech command classifier, text models (Universal Sentence Encoder, toxicity scoring), and a general-purpose KNN classifier utility.Can I try a demo before installing anything?Each model entry links to a live demo and source code so you can see it in action before adding the npm package to your own project.
Each model entry links to a live demo and source code so you can see it in action before adding the npm package to your own project.How do I propose a new model for this repo?File a GitHub issue on the main tfjs repository to gauge interest; the maintainers prioritize models that complement the existing set and serve as building blocks for other apps.
File a GitHub issue on the main tfjs repository to gauge interest; the maintainers prioritize models that complement the existing set and serve as building blocks for other apps.How do I run the tests for a model?Run 'yarn test' inside a specific model's directory, or 'yarn presubmit' from the repo root to test all models.
Run 'yarn test' inside a specific model's directory, or 'yarn presubmit' from the repo root to test all models.What are similar projects?transformers.js, X-AnyLabeling, cvat, vision_transformer, ultralytics, face-api.js, framefields, fiftyone
transformers.js, X-AnyLabeling, cvat, vision_transformer, ultralytics, face-api.js, framefields, fiftyoneWhich topics is it in?transfer-learning, pose-estimation, face-detection, image-classification, object-detection, pretrained-models, nlp, computer-vision, sdk, browser, deep-learning, machine-learning
transfer-learning, pose-estimation, face-detection, image-classification, object-detection, pretrained-models, nlp, computer-vision, sdk, browser, deep-learning, machine-learning