For Researchers —

Explore and extend models
from the latest
cutting edge research

Deeplabv3-ResNet101

DeepLabV3 model with a ResNet-101 backbone

Transformer (NMT)

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

WaveGlow

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

ResNext WSL

ResNext models trained with billion scale weakly-supervised data.

DCGAN on FashionGen

A simple generative image model for 64x64 images

Progressive Growing of GANs (PGAN)

High-quality image generation of fashion, celebrity faces

Semi-supervised and semi-weakly supervised ImageNet Models

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

PyTorch-Transformers

PyTorch implementations of popular NLP Transformers

U-Net for brain MRI

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

SSD

Single Shot MultiBox Detector model for object detection

Tacotron 2

The Tacotron 2 model for generating mel spectrograms from text

RoBERTa

A Robustly Optimized BERT Pretraining Approach

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.

Densenet

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

FCN-ResNet101

Fully-Convolutional Network model with a ResNet-101 backbone

GoogLeNet

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

HarDNet

Harmonic DenseNet pre-trained on ImageNet

Inception_v3

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

MobileNet v2

Efficient networks optimized for speed and memory, with residual blocks

ProxylessNAS

Proxylessly specialize CNN architectures for different hardware platforms.

ResNet

Deep residual networks pre-trained on ImageNet

ResNext

Next generation ResNets, more efficient and accurate

ShuffleNet v2

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

SqueezeNet

Alexnet-level accuracy with 50x fewer parameters.

vgg-nets

Award winning ConvNets from 2014 Imagenet ILSVRC challenge

Wide ResNet

Wide Residual Networks
All Research Models (26)

How it works —

Publishing Models

PyTorch Hub supports publishing pre-trained models (model definitions and pre-trained weights) to a GitHub repository by adding a simple hubconf.py file.

Loading models

Users can load pre-trained models using torch.hub.load() API.

Here’s an example showing how to load the resnet18 entrypoint from the pytorch/vision repo.

model = torch.hub.load('pytorch/vision', 'resnet18', pretrained=True)