explaingit

nvidia/fastphotostyle

★ 11,181PythonAudience · researcherComplexity · 3/5LicenseSetup · moderate

TLDR

NVIDIA Python tool that transfers the colors, tones, and textures of one photograph onto another to produce photorealistic results, developed with UC Merced and presented at ECCV 2018.

Mindmap

mindmap
  root((fastphotostyle))
    What it does
      Photo style transfer
      Photorealistic output
      Blends two images
    Inputs
      Content photo
      Style photo
    Tech Stack
      Python
      PyTorch 0.4.0
    Use Cases
      Photography editing
      Creative projects
      Research reproduction
    Audience
      AI researchers
      Computer vision devs
Click or tap to explore — scroll the page freely

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

Things people build with this

USE CASE 1

Transfer the golden-hour lighting and color palette from a reference photo onto your own daytime photograph.

USE CASE 2

Apply the visual style of a specific movie still or artwork to a set of your own photos for creative projects.

USE CASE 3

Reproduce or extend the photorealistic style transfer research from the ECCV 2018 paper in your own experiments.

Tech stack

PythonPyTorch

Getting it running

Difficulty · moderate Time to first run · 1h+

Requires PyTorch 0.4.0 specifically, a separate older release exists for PyTorch 0.3.0, and modern PyTorch versions are not supported.

Free for non-commercial research use with attribution, any work built on it must carry the same non-commercial license terms.

In plain English

FastPhotoStyle is a Python project from NVIDIA that transfers the visual style of one photograph onto another photograph. You give it two images: a content photo (the scene you want to keep) and a style photo (the look you want to borrow), and it produces a result that blends the content of the first with the colors, tones, and textures of the second. The goal is photorealistic output, meaning the result should look like a real photo rather than an obvious digital painting. The technique was developed by researchers from NVIDIA and UC Merced and published at the ECCV 2018 conference. The project README is brief and points to a separate tutorial file for three different ways to use the algorithm. The code is written in Python and was built using PyTorch, a widely used machine learning framework. The README notes it was updated to work with PyTorch 0.4.0, with an older release available for users on version 0.3.0. The license is CC BY-NC-SA 4.0, which allows non-commercial use with attribution and requires that any work built on top of it carries the same license terms. Commercial use is not permitted under this license.

Copy-paste prompts

Prompt 1
Using NVIDIA's FastPhotoStyle with PyTorch 0.4.0, transfer the style of style.jpg onto content.jpg and save the result, show me the exact command from the tutorial.
Prompt 2
I have a content photo and a style photo. Walk me through installing FastPhotoStyle's dependencies and running the style transfer pipeline end to end.
Prompt 3
How does FastPhotoStyle differ from neural style transfer, what makes the output photorealistic rather than painterly? Explain based on the ECCV 2018 paper it references.
Prompt 4
I want to batch-process 50 content photos with the same style reference using FastPhotoStyle, how do I write a loop around the main command?
Open on GitHub → Explain another repo

← nvidia on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.