Skip to main content

PyTorch Hub For Researchers

Explore and extend models from the latest cutting edge research.

Discover and publish models to a pre-trained model repository designed for research exploration. Check out the models for Researchers, or learn How It WorksContribute Models.

*This is a beta release – we will be collecting feedback and improving the PyTorch Hub over the coming months.

  • Reset

PyTorch-Transformers

PyTorch implementations of popular NLP Transformers

144.7k

YOLOv5

Ultralytics YOLOv5 🚀 for object detection, instance segmentation and image classification.

53.9k

Transformer (NMT)

Transformer models for English-French and English-German translation.

31.5k

RoBERTa

A Robustly Optimized BERT Pretraining Approach

31.5k

Deeplabv3

DeepLabV3 models with ResNet-50, ResNet-101 and MobileNet-V3 backbones

16.8k

AlexNet

The 2012 ImageNet winner achieved a top-5 error of 15.3%, more than 10.8 percentage points lower than that of the runner up.

16.8k

SqueezeNet

Alexnet-level accuracy with 50x fewer parameters.

16.8k

Densenet

Dense Convolutional Network (DenseNet), connects each layer to every other layer in a feed-forward fashion.

16.8k

vgg-nets

Award winning ConvNets from 2014 ImageNet ILSVRC challenge

16.8k

FCN

Fully-Convolutional Network model with ResNet-50 and ResNet-101 backbones

16.8k

Wide ResNet

Wide Residual Networks

16.8k

GoogLeNet

GoogLeNet was based on a deep convolutional neural network architecture codenamed “Inception” which won ImageNet 2014.

16.8k

Inception_v3

Also called GoogleNetv3, a famous ConvNet trained on ImageNet from 2015

16.8k

MobileNet v2

Efficient networks optimized for speed and memory, with residual blocks

16.8k

ResNet

Deep residual networks pre-trained on ImageNet

16.8k

ResNext

Next generation ResNets, more efficient and accurate

16.8k

ShuffleNet v2

An efficient ConvNet optimized for speed and memory, pre-trained on ImageNet

16.8k

FastPitch 2

The FastPitch model for generating mel spectrograms from text

14.3k

GPUNet

GPUNet is a new family of Convolutional Neural Networks designed to max out the performance of NVIDIA GPU and TensorRT.

14.3k

EfficientNet

EfficientNets are a family of image classification models, which achieve state-of-the-art accuracy, being an order-of-magnitude smaller and faster. Trained with mixed precision using Tensor Cores.

14.3k

HiFi GAN

The HiFi GAN model for generating waveforms from mel spectrograms

14.3k

ResNet50

ResNet50 model trained with mixed precision using Tensor Cores.

14.3k

ResNeXt101

ResNet with bottleneck 3×3 Convolutions substituted by 3×3 Grouped Convolutions, trained with mixed precision using Tensor Cores.

14.3k

SE-ResNeXt101

ResNeXt with Squeeze-and-Excitation module added, trained with mixed precision using Tensor Cores.

14.3k

SSD

Single Shot MultiBox Detector model for object detection

14.3k

Tacotron 2

The Tacotron 2 model for generating mel spectrograms from text

14.3k

WaveGlow

WaveGlow model for generating speech from mel spectrograms (generated by Tacotron2)

14.3k

Silero Voice Activity Detector

Pre-trained Voice Activity Detector

5.9k

Silero Speech-To-Text Models

A set of compact enterprise-grade pre-trained STT Models for multiple languages.

5.3k

Silero Text-To-Speech Models

A set of compact enterprise-grade pre-trained TTS Models for multiple languages

5.3k

MiDaS

MiDaS models for computing relative depth from a single image.

4.9k

GhostNet

Efficient networks by generating more features from cheap operations

4.2k

SNNMLP

Brain-inspired Multilayer Perceptron with Spiking Neurons

4.2k

3D ResNet

Resnet Style Video classification networks pretrained on the Kinetics 400 dataset

3.4k

SlowFast

SlowFast networks pretrained on the Kinetics 400 dataset

3.4k

X3D

X3D networks pretrained on the Kinetics 400 dataset

3.4k

ResNeSt

A new ResNet variant.

3.3k

YOLOP

YOLOP pretrained on the BDD100K dataset

2.0k

Once-for-All

Once-for-all (OFA) decouples training and search, and achieves efficient inference across various edge devices and resource constraints.

1.9k

DCGAN on FashionGen

A simple generative image model for 64×64 images

1.6k

Progressive Growing of GANs (PGAN)

High-quality image generation of fashion, celebrity faces

1.6k

ProxylessNAS

Proxylessly specialize CNN architectures for different hardware platforms.

1.4k

Open-Unmix

Reference implementation for music source separation

1.4k

IBN-Net

Networks with domain/appearance invariance

803

U-Net for brain MRI

U-Net with batch normalization for biomedical image segmentation with pretrained weights for abnormality segmentation in brain MRI

753

MEAL_V2

Boosting Tiny and Efficient Models using Knowledge Distillation.

694

HybridNets

HybridNets – End2End Perception Network

634

ResNext WSL

ResNext models trained with billion scale weakly-supervised data.

602

HarDNet

Harmonic DenseNet pre-trained on ImageNet

370

Semi-supervised and semi-weakly supervised ImageNet Models

ResNet and ResNext models introduced in the “Billion scale semi-supervised learning for image classification” paper

243

SimpleNet

Lets Keep it simple, Using simple architectures to outperform deeper and more complex architectures

52

ntsnet

classify birds using this fine-grained image classifier

34