Edge-AI Model Zoos

A curated list of Model Zoos & Hubs where you can find production-ready and optimized models for resource-constrained devices.

Table of Contents

1. Model Zoos & Hubs
2. Model by Domain & Use Case
3. Resources
   • How to choose the best model for Edge AI application
   • Edge AI Technical Guide for Developers and Practitioners

1. Model Zoos & Hubs

Back to Table of Contents

Model Zoo	Description	Links
Edge AI Labs Model Zoo	A collection of pre-trained, optimized models for low-power devices.	EdgeAI Labs
Edge Impulse Model Zoo	A repository of models optimized for edge devices.	Edge Impulse Model Zoo
ONNX Model Zoo	A collection of pre-trained, state-of-the-art models in the ONNX format.	ONNX Model Zoo
NVIDIA Pretrained AI Models (NGC + TAO)	Accelerate AI development with world-class customizable pretrained models from NVIDIA.	- NVIDIA Pretrained AI Models - Main - NGC Model Catalog - TAO Model Zoo
OpenVINO Model Zoo	A collection of pre-trained models ready for use with Intel's OpenVINO toolkit.	OpenVINO Model Zoo
Qualcomm Models Zoo	A collection of AI models from Qualcomm.	Qualcomm Models Zoo
LiteRT Pre-trained models	Pre-trained models optimized for Google's Lite Runtime.	LiteRT Pre-trained Models
Keras Applications	Pre-trained models for Keras applications	Keras Pre-trained Models
MediaPipe	Framework for building multimodal applied machine learning pipelines.	MediaPipe
TensorFlow Model Garden	A repository with a collection of TensorFlow models.	TensorFlow Model Garden
Pytorch Model Zoo	A hub for pre-trained models on PyTorch framework.	Pytorch Model Zoo
stm32ai-modelzoo	AI Model Zoo for STM32 microcontroller devices.	stm32ai-modelzoo
Model Zoo	A collection of pre-trained models for various machine learning tasks.	Model Zoo
Hugging Face Models	A collection of pre-trained models for various machine learning tasks.	Hugging Face Models
Papers with Code	A repository that links academic papers to their respective code and models.	Papers with Code
MXNet Model Zoo	A collection of pre-trained models for the Apache MXNet framework.	MXNet Model Zoo
Deci’s Model Zoo	A curated list of high-performance deep learning models.	Deci’s Model Zoo
Jetson Model Zoo and Community Projects	NVIDIA's collection of models and projects for Jetson platform.	Jetson Model Zoo and Community Projects
Magenta	Models for music and art generation from Google's Magenta project.	Magenta
Awesome-CoreML-Models Public	A collection of CoreML models for iOS developers.	Awesome-CoreML-Models Public
Pinto Models	A variety of models for computer vision tasks.	Pinto Models
Baidu AI Open Model Zoo	Baidu's collection of AI models.	Baidu AI Open Model Zoo
Hailo Model Zoo	A set of models optimized for Hailo's AI processors.	Hailo Model Zoo

2. Model by Domain & Use Case

Back to Table of Contents

This is a non-exhaustive selection of models from several platforms listed in Section 1, ranged into six domains and a variety of tasks, with a focus on efficiency and real-world applications.

Domain	Task	Model	Description	Reference
Computer Vision	Object Detection	yolov8_det	Object detection for edge devices	YOLOv8 on GitHub
	Image Classification	mobilenet_v3_small	Lightweight image classification	MobileNetV3 on TensorFlow Hub
	Semantic Segmentation	deeplabv3_resnet50	Semantic image segmentation	DeepLabV3 on TensorFlow Hub
	Instance Segmentation	yolov8_seg	Object detection and segmentation	YOLOv8 on GitHub
	Object Tracking	DeepSort	Real-time object tracking	DeepSort on GitHub
	Pose Estimation	openpose	Human pose estimation	OpenPose on GitHub
	Facial Recognition	mediapipe_face	Face detection and recognition	MediaPipe Face on Google AI
	Optical Character Recognition	trocr	Text recognition in images	TrOCR on Hugging Face
	Video Classification	resnet_2plus1d	Video classification for action recognition	ResNet-2+1D on PyTorch Hub
	Video Classification	resnet_3d	3D CNN for video classification	ResNet-3D on PyTorch Hub
Audio Processing	Speech-to-Text	distil-whisper	Lightweight speech recognition model	Distil-Whisper on Hugging Face
	Sound Classification	audio-spectrogram-transformer	Transformer for audio classification	AST on Hugging Face
	Voice Activity Detection	silero-vad	Voice activity detection for edge devices	Silero VAD on GitHub
	Acoustic Scene Classification	panns	Audio tagging and scene classification	PANNS on GitHub
	Speaker Diarization	pyannote-audio	Speaker diarization and segmentation	PyAnnote on Hugging Face
	Speech Recognition	wav2vec2	Self-supervised speech representation learning	Wav2vec2 on Hugging Face
	-	-	-	-
	-	-	-	-
	-	-	-	-
	-	-	-	-
Time Series	Predictive Maintenance	tsmixer	Time-series forecasting for maintenance	TimesFM on GitHub
	Anomaly Detection	IsolationForest	Anomaly detection in time-series data	IsolationForest on Scikit-learn
	Forecasting	informer	Transformer-based time-series forecasting	Informer on GitHub
	Time-Series Classification	rocket	Efficient time-series classification	ROCKET on GitHub
	Image Super-Resolution	real_esrgan_x4plus	Image super-resolution for temporal data	Real-ESRGAN on GitHub
	Image Inpainting	lama_dilated	Image inpainting for time-series analysis	LaMa on GitHub
	-	-	-	-
	-	-	-	-
	-	-	-	-
	-	-	-	-
NLP	Speech Recognition	Whisper	General-purpose speech recognition model	Whisper on Hugging Face
	Keyword Spotting	silero-kws	Wake word detection for edge devices	Silero Models on GitHub
	Text Classification	distilbert	Lightweight transformer for text classification	DistilBERT on Hugging Face
	Named Entity Recognition	bert-ner	NER for entity extraction	BERT-NER on Hugging Face
	Question Answering	mobilebert	Lightweight QA model for edge	MobileBERT on Hugging Face
	Text Summarization	bart	Text summarization for short texts	BART on Hugging Face
	-	-	-	-
	-	-	-	-
	-	-	-	-
	-	-	-	-
Generative AI	Image Generation & Synthesis	ControlNet	Fine control over image generation	ControlNet on GitHub
		Stable Diffusion	Text-to-image generation	Stable Diffusion on Hugging Face
		stylegan2	Image generation	StyleGAN2 on GitHub
		Flux.1-schnell	Fast text-to-image generation	Flux.1 on Hugging Face, Awesome-Smol
		Reve	Image generation with advanced text rendering	Reve on Hugging Face
		-	-	-
		-	-	-
	Small Language Model (SLM)	SmolLM2-1.7B	Small language model for efficient text generation	SmolLM2 on Hugging Face, Awesome-Smol
		Gemma 2	Lightweight open model for text generation	Gemma 2 on Hugging Face, Awesome-Smol
		Phi-3.5-mini	Small language model with strong reasoning	Phi-3.5-mini on Hugging Face, Awesome-Smol
		Qwen2.5-1.5B	Efficient language model for instruction following	Qwen2.5 on Hugging Face, Awesome-Smol
		Mixtral-8x22B	Sparse mixture of experts for text generation	Mixtral on Hugging Face
		-	-	-
		-	-	-
	Multimodality	SmolVLM-256M	Smallest vision-language model for image understanding	SmolVLM-256M on Hugging Face, Awesome-Smol
		SmolVLM-500M	Vision-language model for image and text tasks	SmolVLM-500M on Hugging Face, Awesome-Smol
		BakLLaVA-1	Multimodal model for text and image tasks	BakLLaVA-1 on Hugging Face, Awesome-Smol
		PaliGemma	Vision-language model for multimodal tasks	PaliGemma on Hugging Face
		Seed1.5-VL	Vision-language model with strong multimodal performance	Seed1.5-VL on Hugging Face
		-	-	-
		-	-	-
Misc	Sensor Fusion	mediapipe_pose	Human pose estimation using sensor data	MediaPipe Pose on Google AI
	Activity Recognition	har-cnn	Human activity recognition from sensor data	HAR-CNN on GitHub
	Contextual Awareness	SmolVLM-256M	Multimodal model for environment understanding	SmolVLM-256M on Hugging Face, Awesome-Smol
	Network Anomaly Detection	LOF	Local outlier factor for network anomalies	LOF on Scikit-learn
	Device Behavior Anomaly	Autoencoder	Anomaly detection for device behavior	Keras Autoencoder
	Sensor Data Anomaly	OC-SVM	One-class SVM for sensor data anomalies	OneClassSVM on Scikit-learn
	On-device Control Systems	TD3	Twin Delayed DDPG for control systems	TD3 on GitHub
	Various	pinecone	Vector database	-
	Various	weaviate-c2	Vector database	-
	Various	upstage	Various models	Awesome-Smol

Resources

How to choose the best model for an Edge AI application

Selecting the right model for edge deployment is critical for balancing performance, accuracy and efficiency.

Why It Matters

• Efficiency: Edge devices (e.g., IoT, mobile, embedded systems) have limited compute, memory, and power.
• Performance: Real-time applications (e.g., autonomous drones, smart cameras) demand low latency and high accuracy.
• Scalability: The right model ensures cost-effective deployment across devices.

Key Criteria

1. Task Requirements: Match the model to your application (e.g., vision, audio, multimodal).
2. Hardware Constraints: Consider compute (OPS), memory (MB), and energy (mWh) limits of your device.
3. Performance Goals: Balance accuracy, latency, and throughput for your use case.
4. Deployment Ease: Check compatibility with frameworks (e.g., TensorFlow Lite, ONNX).

Next Steps: Once you’ve shortlisted a model, use the Edge AI Benchmarking Guide to profile and optimize the model performance.

Edge AI Technical Guide for Developers and Practitioners

• Edge AI Engineering
• Edge AI Technical Guide
• Edge AI End-to-End Stack
• Edge AI Deployment Stack
• Edge AI Optimization Stack
• Edge AI Frameworks
• Edge AI Model Zoos
• Edge AI Platforms
• Edge AI Benchmarking
• Edge AI Ecosystem
• Edge AI Books
• Edge AI Blog
• Edge AI Papers

Back to Table of Contents