An implementation of 'Deep Bilateral Learning for Real-Time Image Enhancement', SIGGRAPH 2017.
google/hdrnet is ranked 12,968 with 877 stars.
Same-category comparison
google/hdrnet is ranked 12,968 among these 8 peers; yule-BUAA/MergeLM is ranked 12,965.
Same-category comparison| Name | Rank | Metrics |
|---|
| dmlc/dmlc-core | 12,959 | - Language:
- C++
- Stars:
- 878
- Forks:
- 525
- Description:
- A common bricks library for building scalable and portable distributed machine learning.
|
|---|
| lucidrains/RETRO-pytorch | 12,961 | - Language:
- Python
- Stars:
- 878
- Forks:
- 108
- Description:
- Implementation of RETRO, Deepmind's Retrieval based Attention net, in Pytorch.
|
|---|
| mystic123/tensorflow-yolo-v3 | 12,962 | - Language:
- Python
- Stars:
- 878
- Forks:
- 343
- Description:
- Implementation of YOLO v3 object detector in Tensorflow (TF-Slim).
|
|---|
| yule-BUAA/MergeLM | 12,965 | - Language:
- Python
- Stars:
- 878
- Forks:
- 53
- Description:
- Codebase for Merging Language Models (ICML 2024).
|
|---|
| EpistasisLab/pmlb | 12,973 | - Language:
- Python
- Stars:
- 876
- Forks:
- 144
- Description:
- PMLB: A large, curated repository of benchmark datasets for evaluating supervised machine learning algorithms.
|
|---|
| WooooDyy/AgentGym-RL | 12,975 | - Language:
- Python
- Stars:
- 876
- Forks:
- 89
- Description:
- Code and implementations for the paper "AgentGym-RL: Training LLM Agents for Long-Horizon Decision Making through Multi-Turn Reinforcement Learning" by Zhiheng Xi et al.
|
|---|
| brownvc/R3GAN | 12,976 | - Language:
- Python
- Stars:
- 876
- Forks:
- 45
- Description:
- Code for NeurIPS 2024 paper - The GAN is dead; long live the GAN. A Modern Baseline GAN - by Huang et al.
|
|---|
| polarisZhao/PFLD-pytorch | 12,984 | - Language:
- Python
- Stars:
- 876
- Forks:
- 205
- Description:
- PFLD pytorch Implementation.
|
|---|
Key parameters and specifications
Key parameters and specifications| Parameter | Value |
|---|
| Language | Python |
|---|
| Category | Model Development |
|---|
| Subcategory | Architecture |
|---|
| Keywords | Image production |
|---|
| Stars | 877 |
|---|
| Forks | 319 |
|---|
| Contributors | 4 |
|---|
| Created date | 2016-09-12 |
|---|
| Updated date | 2026-10-02 |
|---|
| Stars in 1 day | 0 |
|---|
| Stars in 7 days | 0 |
|---|
| Rank | 12,968 |
|---|
| Previous rank | 12,865 |
|---|
| Rank change | -103 |
|---|
| 7-day stars rank | 13,696 |
|---|
| Previous 7-day stars rank | 13,709 |
|---|
| 7-day stars rank change | 13 |
|---|
| Total stars rank | 12,349 |
|---|
| Previous total stars rank | 12,324 |
|---|
| Total stars rank change | -25 |
|---|
| Country | USA |
|---|
README summary
Deep Bilateral Learning for Real-Time Image Enhancements Siggraph 2017 Visit our Project Page. Hasinoff Fredo Durand Maintained by Michael Gharbi ( ) Tested on Python 2. Disclaimer This is not an official Google product. Setup Dependencies To install the Python dependencies, run: Build Our network requires a custom Tensorflow operator to "slice" in the bilateral grid. To build it, run: To build the benchmarking code, run: Note that the benchmarking code requires a frozen and optimized model. To build the Android
demo, see dedicated section below. Test Run the test suite to make sure the BilateralSlice operator works correctly: Download pretrained models We provide a set of pretrained models. One of these is included in the repo (see ). To download the rest of them run: Usage To train a model, run the following command: Look at for a typical structure of the training data folder. You can monitor the training process using Tensorboard: To run a trained model on a novel
image (or set of images), use: To prepare a model for use on mobile, freeze the graph, and optimize the network: You will need to change the environment variable in and compile the necessary tensorflow command line tools for this (automated in the script). Android prototype We will add it to this repo soon. Known issues and limitations The BilateralSliceApply operation is GPU only at this point. We do not plan on releasing a CPU implementation. The provided pre-trained
models were updated from an older version and might slightly differ from the models used for evaluation in the paper. The pre-trained HDR+ model expects as input a specially formatted 16-bit linear input. Apply lens shading correction (aka vignetting correction). Our Android demo approximates this by undoing the RGB->YUV conversion and white balance, and tone mapping performed by the Qualcomm SOC. It results in slightly different colors than that on the test set. If you run our
HDR+ model on an sRGB input, it may produce uncanny colors.
Frequently asked questions
How many stars and forks are recorded?
google/hdrnet has 877 stars and 319 forks.
What is the repository rank?
google/hdrnet is ranked 12,968.
What language and category are recorded?
google/hdrnet has 877 stars, uses Python and belongs to Model Development.
Which repository is closest in rank?
google/hdrnet is compared with yule-BUAA/MergeLM, ranked 12,965 with 878 stars.
Deep Bilateral Learning for Real-Time Image Enhancements
Siggraph 2017
Visit our Project Page.
Michael Gharbi
Jiawen Chen
Jonathan T. Barron
Samuel W. Hasinoff
Fredo Durand
Maintained by Michael Gharbi ()
Tested on Python 2.7, Ubuntu 14.0, gcc-4.8.
Disclaimer
This is not an official Google product.
Setup
Dependencies
To install the Python dependencies, run:
cd hdrnet
pip install -r requirements.txt
Build
Our network requires a custom Tensorflow operator to "slice" in the bilateral grid.
To build it, run:
cd hdrnet
make
To build the benchmarking code, run:
cd benchmark
make
Note that the benchmarking code requires a frozen and optimized model. Use
hdrnet/bin/scripts/optimize_graph.sh and hdrnet/bin/freeze.py to produce these.
To build the Android demo, see dedicated section below.
Test
Run the test suite to make sure the BilateralSlice operator works correctly:
cd hdrnet
py.test test
Download pretrained models
We provide a set of pretrained models. One of these is included in the repo
(see pretrained_models/local_laplacian_sample). To download the rest of them
run:
cd pretrained_models
./download.py
Usage
To train a model, run the following command:
./hdrnet/bin/train.py
Look at sample_data/identity/ for a typical structure of the training data folder.
You can monitor the training process using Tensorboard:
tensorboard --logdir
To run a trained model on a novel image (or set of images), use:
./hdrnet/bin/run.py
To prepare a model for use on mobile, freeze the graph, and optimize the network:
./hdrnet/bin/freeze_graph.py
./hdrnet/bin/scripts/optimize_graph.sh
You will need to change the ${TF_BASE} environment variable in ./hdrnet/bin/scripts/optimize_graph.sh
and compile the necessary tensorflow command line tools for this (automated in the script).
Android prototype
We will add it to this repo soon.
Known issues and limitations
-
The BilateralSliceApply operation is GPU only at this point. We do not plan on releasing a CPU implementation.
-
The provided pre-trained models were updated from an older version and might slightly differ from the models used for evaluation in the paper.
-
The pre-trained HDR+ model expects as input a specially formatted 16-bit linear input. In summary, starting from Bayer RAW:
- Subtract black level.
- Apply white balance channel gains.
- Demosaic to RGB.
- Apply lens shading correction (aka vignetting correction).
Our Android demo approximates this by undoing the RGB->YUV conversion and
white balance, and tone mapping performed by the Qualcomm SOC. It results in slightly different colors than that on the test set. If you run our HDR+ model on an sRGB input, it may produce uncanny colors.