OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Image & Video AI

Exposure

Exposure is an open-source Python research framework for learning photo post-processing from unpaired image datasets with GANs, reinforcement learning and a differentiable photo-editing model. It includes a pretrained model and code for training and evaluating custom models.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Exposure — OpenViST project profile in Image & Video AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Exposure when you want a research-code workflow for learning photo post-processing from image collections and evaluating a pretrained or custom model, especially with linear, high-bit-depth photo inputs.

Who it is for

  • Computational photography and image-processing researchers exploring learned photo editing.
  • Developers who want to train or evaluate image post-processing models using the repository’s documented workflows.

Documented uses

  • Apply the included pretrained model to sample TIFF images.
  • Train a model with the MIT-Adobe FiveK dataset and example configuration.
  • Explore training on another dataset using the repository’s Sintel configuration.
Before you choose

Requirements & limits

The project is a public GitHub repository with an MIT license. Its documented setup requires Python 3 and TensorFlow; the README reports testing on Ubuntu 16.04 and Arch Linux, while OS X is untested.

  • The included pretrained model is available through a recursive repository clone because it is provided as a submodule.
  • The FiveK fetch script downloads a downsampled, augmented pack of about 2.4 GB; working from original RAW files requires a separate Lightroom preprocessing workflow.
  • Image characteristics matter: low-bit-depth or gamma-encoded JPEG and PNG inputs can yield suboptimal output or posterization.
  • Dropout supports one-to-many output behavior, so repeated runs on the same image may differ.
  • The README presents the project as a proof-of-concept and warns that inputs unlike the training data can produce suboptimal results.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Apply the included pretrained model to sample TIFF images

Evaluate the repository’s pretrained model on TIFF images in the sample-input folder and write the results to outputs/.

Before you start

  • Python 3 and TensorFlow.
  • The repository’s listed Python packages: tensorflow-gpu, opencv-python, tifffile and scikit-image.
  • A recursive clone so the pretrained-model submodule is included.

Steps

  1. Install the listed Python packages.

  2. Clone the repository recursively and change into its directory.

  3. Run the command below.

    Documented command
    python3 evaluate.py example pretrained models/sample_inputs/*.tif

  4. Find generated results in outputs/.

Expected outcome

Evaluation results are generated in the outputs/ directory.

Guide limitations

  • The README describes a research prototype; input differences from training data may lead to suboptimal results.
  • The pretrained model was trained on linearized ProPhotoRGB data, and JPEG or PNG inputs may show posterization or other quality issues.
  • The README reports testing on Ubuntu 16.04 and Arch Linux; OS X is mentioned as possibly working but untested.
Documentation 1 ↗
Source-based practical guide

Train and evaluate a model with the MIT-Adobe FiveK data

Fetch the documented downsampled FiveK image pack, train using the example configuration, and evaluate the resulting model.

Before you start

  • Python 3 and TensorFlow, plus tensorflow-gpu, opencv-python, tifffile and scikit-image.
  • A recursive clone of the repository.
  • Approximately 2.4 GB for the downsampled and augmented image pack downloaded by the fetch script.

Steps

  1. Run the command below to set up the FiveK data pack.

    Documented command
    python3 fetch_fivek.py

  2. Run the command below; this loads config_example.py and creates a model folder at models/example/test.

    Documented command
    python3 train.py example test

  3. Run the command below.

    Documented command
    python3 evaluate.py example test models/sample_inputs/*.tif

  4. Find generated results in outputs/; training progress images are written under the model folder.

Expected outcome

A model trained with the example configuration is saved under `models/example/test`, and evaluation results are generated in `outputs/`.

Guide limitations

  • The README estimates about 100 minutes of training on a GTX 1080 Ti; this is a reported estimate, not a general runtime guarantee.
  • The original FiveK dataset is described as roughly 50 GB; preprocessing RAW files requires Adobe Lightroom if preparing the data independently.
  • The framework is identified by its authors as a proof-of-concept prototype, not a finished production product.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
17 October 2026
Original website title
yuanming-hu/exposure · Official homepage title · retained in its source language
How we review sourcesSuggest a correction