High-fidelity performance metrics for generative models in PyTorch

Overview

High-fidelity performance metrics for generative models in PyTorch

Documentation Status TestStatus PyPiVersion PyPiDownloads Twitter Follow

This repository provides precise, efficient, and extensible implementations of the popular metrics for generative model evaluation, including:

  • Inception Score (ISC)
  • Fréchet Inception Distance (FID)
  • Kernel Inception Distance (KID)
  • Perceptual Path Length (PPL)

Precision: Unlike many other reimplementations, the values produced by torch-fidelity match reference implementations up to machine precision. This allows using torch-fidelity for reporting metrics in papers instead of scattered and slow reference implementations. Read more about precision

Efficiency: Feature sharing between different metrics saves recomputation time, and an additional caching level avoids recomputing features and statistics whenever possible. High efficiency allows using torch-fidelity in the training loop, for example at the end of every epoch. Read more about efficiency

Extensibility: Going beyond 2D image generation is easy due to high modularity and abstraction of the metrics from input data, models, and feature extractors. For example, one can swap out InceptionV3 feature extractor for a one accepting 3D scan volumes, such as used in MRI. Read more about extensibility

TLDR; fast and reliable GAN evaluation in PyTorch

Installation

pip install torch-fidelity

See also: Installing the latest GitHub code

Usage Examples with Command Line

Below are three examples of using torch-fidelity to evaluate metrics from the command line. See more examples in the documentation.

Simple

Inception Score of CIFAR-10 training split:

> fidelity --gpu 0 --isc --input1 cifar10-train

inception_score_mean: 11.23678
inception_score_std: 0.09514061

Medium

Inception Score of a directory of images stored in ~/images/:

> fidelity --gpu 0 --isc --input1 ~/images/

Pro

Efficient computation of ISC and PPL for input1, and FID and KID between a generative model stored in ~/generator.onnx and CIFAR-10 training split:

> fidelity \
  --gpu 0 \
  --isc \
  --fid \
  --kid \
  --ppl \
  --input1 ~/generator.onnx \ 
  --input1-model-z-type normal \
  --input1-model-z-size 128 \
  --input1-model-num-samples 50000 \ 
  --input2 cifar10-train 

See also: Other usage examples

Quick Start with Python API

When it comes to tracking the performance of generative models as they train, evaluating metrics after every epoch becomes prohibitively expensive due to long computation times. torch_fidelity tackles this problem by making full use of caching to avoid recomputing common features and per-metric statistics whenever possible. Computing all metrics for 50000 32x32 generated images and cifar10-train takes only 2 min 26 seconds on NVIDIA P100 GPU, compared to >10 min if using original codebases. Thus, computing metrics 20 times over the whole training cycle makes overall training time just one hour longer.

In the following example, assume unconditional image generation setting with CIFAR-10, and the generative model generator, which takes a 128-dimensional standard normal noise vector.

First, import the module:

import torch_fidelity

Add the following lines at the end of epoch evaluation:

wrapped_generator = torch_fidelity.GenerativeModelModuleWrapper(generator, 128, 'normal', 0)

metrics_dict = torch_fidelity.calculate_metrics(
    input1=wrapped_generator, 
    input2='cifar10-train', 
    cuda=True, 
    isc=True, 
    fid=True, 
    kid=True, 
    verbose=False,
)

The resulting dictionary with computed metrics can logged directly to tensorboard, wandb, or console:

print(metrics_dict)

Output:

{
    'inception_score_mean': 11.23678, 
    'inception_score_std': 0.09514061, 
    'frechet_inception_distance': 18.12198,
    'kernel_inception_distance_mean': 0.01369556, 
    'kernel_inception_distance_std': 0.001310059
}

See also: Full API reference

Example of Integration with the Training Loop

Refer to sngan_cifar10.py for a complete training example.

Evolution of fixed generator latents in the example:

Evolution of fixed generator latents

A generator checkpoint resulting from training the example can be downloaded here.

Citation

Citation is recommended to reinforce the evaluation protocol in works relying on torch-fidelity. To ensure reproducibility when citing this repository, use the following BibTeX:

@misc{obukhov2020torchfidelity,
  author={Anton Obukhov and Maximilian Seitzer and Po-Wei Wu and Semen Zhydenko and Jonathan Kyl and Elvis Yu-Jing Lin},
  year=2020,
  title={High-fidelity performance metrics for generative models in PyTorch},
  url={https://github.com/toshas/torch-fidelity},
  publisher={Zenodo},
  version={v0.3.0},
  doi={10.5281/zenodo.4957738},
  note={Version: 0.3.0, DOI: 10.5281/zenodo.4957738}
}
Owner
Vikram Voleti
PhD student at Mila, University of Montreal
Vikram Voleti
Kaldi-compatible feature extraction with PyTorch, supporting CUDA, batch processing, chunk processing, and autograd

Kaldi-compatible feature extraction with PyTorch, supporting CUDA, batch processing, chunk processing, and autograd

Fangjun Kuang 119 Jan 03, 2023
Pretrained EfficientNet, EfficientNet-Lite, MixNet, MobileNetV3 / V2, MNASNet A1 and B1, FBNet, Single-Path NAS

(Generic) EfficientNets for PyTorch A 'generic' implementation of EfficientNet, MixNet, MobileNetV3, etc. that covers most of the compute/parameter ef

Ross Wightman 1.5k Jan 01, 2023
A PyTorch implementation of EfficientNet

EfficientNet PyTorch Quickstart Install with pip install efficientnet_pytorch and load a pretrained EfficientNet with: from efficientnet_pytorch impor

Luke Melas-Kyriazi 7.2k Jan 06, 2023
A PyTorch implementation of Learning to learn by gradient descent by gradient descent

Intro PyTorch implementation of Learning to learn by gradient descent by gradient descent. Run python main.py TODO Initial implementation Toy data LST

Ilya Kostrikov 300 Dec 11, 2022
3D-RETR: End-to-End Single and Multi-View3D Reconstruction with Transformers

3D-RETR: End-to-End Single and Multi-View 3D Reconstruction with Transformers (BMVC 2021) Zai Shi*, Zhao Meng*, Yiran Xing, Yunpu Ma, Roger Wattenhofe

Zai Shi 36 Dec 21, 2022
This is an differentiable pytorch implementation of SIFT patch descriptor.

This is an differentiable pytorch implementation of SIFT patch descriptor. It is very slow for describing one patch, but quite fast for batch. It can

Dmytro Mishkin 150 Dec 24, 2022
PyTorch framework A simple and complete framework for PyTorch, providing a variety of data loading and simple task solutions that are easy to extend and migrate

PyTorch framework A simple and complete framework for PyTorch, providing a variety of data loading and simple task solutions that are easy to extend and migrate

Cong Cai 12 Dec 19, 2021
Pytorch bindings for Fortran

Pytorch bindings for Fortran

Dmitry Alexeev 46 Dec 29, 2022
Use Jax functions in Pytorch with DLPack

Use Jax functions in Pytorch with DLPack

Phil Wang 106 Dec 17, 2022
A PyTorch repo for data loading and utilities to be shared by the PyTorch domain libraries.

A PyTorch repo for data loading and utilities to be shared by the PyTorch domain libraries.

878 Dec 30, 2022
PyTorch implementation of Glow, Generative Flow with Invertible 1x1 Convolutions

glow-pytorch PyTorch implementation of Glow, Generative Flow with Invertible 1x1 Convolutions

Kim Seonghyeon 433 Dec 27, 2022
PyNIF3D is an open-source PyTorch-based library for research on neural implicit functions (NIF)-based 3D geometry representation.

PyNIF3D is an open-source PyTorch-based library for research on neural implicit functions (NIF)-based 3D geometry representation. It aims to accelerate research by providing a modular design that all

Preferred Networks, Inc. 96 Nov 28, 2022
Implementation of LambdaNetworks, a new approach to image recognition that reaches SOTA with less compute

Lambda Networks - Pytorch Implementation of λ Networks, a new approach to image recognition that reaches SOTA on ImageNet. The new method utilizes λ l

Phil Wang 1.5k Jan 07, 2023
Tacotron 2 - PyTorch implementation with faster-than-realtime inference

Tacotron 2 (without wavenet) PyTorch implementation of Natural TTS Synthesis By Conditioning Wavenet On Mel Spectrogram Predictions. This implementati

NVIDIA Corporation 4.1k Jan 03, 2023
Distiller is an open-source Python package for neural network compression research.

Wiki and tutorials | Documentation | Getting Started | Algorithms | Design | FAQ Distiller is an open-source Python package for neural network compres

Intel Labs 4.1k Dec 28, 2022
A tutorial on "Bayesian Compression for Deep Learning" published at NIPS (2017).

Code release for "Bayesian Compression for Deep Learning" In "Bayesian Compression for Deep Learning" we adopt a Bayesian view for the compression of

Karen Ullrich 190 Dec 30, 2022
A Closer Look at Structured Pruning for Neural Network Compression

A Closer Look at Structured Pruning for Neural Network Compression Code used to reproduce experiments in https://arxiv.org/abs/1810.04622. To prune, w

Bayesian and Neural Systems Group 140 Dec 05, 2022
ocaml-torch provides some ocaml bindings for the PyTorch tensor library.

ocaml-torch provides some ocaml bindings for the PyTorch tensor library. This brings to OCaml NumPy-like tensor computations with GPU acceleration and tape-based automatic differentiation.

Laurent Mazare 369 Jan 03, 2023
Fast, general, and tested differentiable structured prediction in PyTorch

Torch-Struct: Structured Prediction Library A library of tested, GPU implementations of core structured prediction algorithms for deep learning applic

HNLP 1.1k Jan 07, 2023
A lightweight wrapper for PyTorch that provides a simple declarative API for context switching between devices, distributed modes, mixed-precision, and PyTorch extensions.

A lightweight wrapper for PyTorch that provides a simple declarative API for context switching between devices, distributed modes, mixed-precision, and PyTorch extensions.

Fidelity Investments 56 Sep 13, 2022