Please refer to new and updated L3C for new preprocessing scripts solving certain issues.
This repo is a public backup of a hobby project.
$ ./prep_clic.sh ./data
$ python train.py configs/ms/cr.cf configs/dl/clic.cf FRLLIC_logdir
- Extra arguments can be passed using -p (global_config.py)
$ python train.py configs/ms/cr.cf configs/dl/clic.cf FRLLIC_logdir -p upsampling=deconv
- For using pretrained model / restoring training
$ python train.py configs/ms/cr.cf configs/dl/clic.cf FRLLIC_logdir --restore 0502_1213 --restore_restart
$ python test.py FRLLIC_logdir 0502_1213 data/CLIC/test
$ python test.py FRLLIC_logdir 0524_0001 data/CLIC/test --sample=samples
- Encode to out.l3c
$ python l3c.py FRLLIC_logdir 0524_0001 enc /path/to/img out.l3c
$ python l3c.py FRLLIC_logdir 0624_2025 enc ./fixedimg256.png out.l3c
- Decode from out.l3c, save to decoded.png
$ python l3c.py FRLLIC_logdir 0524_0001 dec out.l3c decoded.png
$ python l3c.py FRLLIC_logdir 0624_2025 dec out.l3c decoded.png
Thanks to L3C for implementations of EDSR, logistic mixtures, and arithmetic coding (torchac) and dataset preprocessing scripts.
ETH Zurich
CVPR'19 (oral presentation)
We propose the first practical learned lossless image compression system, L3C, and show that it outperforms the popular engineered codecs, PNG, WebP and JPEG 2000. At the core of our method is a fully parallelizable hierarchical probabilistic model for adaptive entropy coding which is optimized end-to-end for the compression task. In contrast to recent autoregressive discrete probabilistic models such as PixelCNN, our method i) models the image distribution jointly with learned auxiliary representations instead of exclusively modeling the image distribution in RGB space, and ii) only requires three forward-passes to predict all pixel probabilities instead of one for each pixel. As a result, L3C obtains over two orders of magnitude speedups when sampling compared to the fastest PixelCNN variant (Multiscale-PixelCNN). Furthermore, we find that learning the auxiliary representation is crucial and outperforms predefined auxiliary representations such as an RGB pyramid significantly.
This is an updated version of the repo. See issue #14 for details.
Clone the repo and create a conda environment as follows:
conda create --name l3c_env python=3.7 pip --yes
conda activate l3c_env
We need PyTorch 1.1, CUDA, and some PIP packages (If you don't have a GPU, remove cudatoolkit=10.0
):
conda install pytorch=1.1 torchvision cudatoolkit=10.0 -c pytorch
pip install -r pip_requirements.txt
To test our entropy coding, you must also install torchac, as described below.
- We tested this code with Python 3.7 and PyTorch 1.1. PyTorch 1.2 is not supported, but progress is tracked in #5.
- The training code also works with PyTorch 0.4, but for testing, we use the
torchac
module, which needs PyTorch 1.0 or newer to build, see below. - The code relies on
tensorboardX==1.2
, even though TensorBoard is now part of PyTorch (since 1.1)
We release the following trained models:
Name | Training Set | ID | Download Model | |
---|---|---|---|---|
Main Model | L3C | Open Images | 0306_0001 |
L3C.tar.gz |
Baseline | RGB Shared | Open Images | 0306_0002 |
RGB_Shared.tar.gz |
Baseline | RGB | Open Images | 0306_0003 |
RGB.tar.gz |
Main Model | L3C | ImageNet32 | 0524_0004 |
L3C_inet32.tar.gz |
Main Model | L3C | ImageNet64 | 0524_0005 |
L3C_inet64.tar.gz |
See Evaluation of Models to learn how to evaluate on a dataset.
To train a model yourself, you have to first prepare the data as shown in Prepare Open Images Train. Then, use one of the following commands, explained in more detail below:
Model | Train with the following flags to train.py |
---|---|
L3C | configs/ms/cr.cf configs/dl/oi.cf log_dir |
RGB Shared | configs/ms/cr_rgb_shared.cf configs/dl/oi.cf log_dir |
RGB | configs/ms/cr_rgb.cf configs/dl/oi.cf log_dir |
L3C ImageNet32 | configs/ms/cr.cf configs/dl/in32.cf -p lr.schedule=exp_0.75_e1 log_dir |
L3C ImageNet64 | configs/ms/cr.cf configs/dl/in64.cf -p lr.schedule=exp_0.75_e1 log_dir |
Each of the released models were trained for around 5 days on a Titan Xp.
Note: We do not provide code for multi-GPU training. To incorporate nn.DataParallel
, the code must be changed
slightly: In net.py
, EncOut
and DecOut
are namedtuple
s, which is not supported by nn.DataParallel
.
To test an experiment, use test.py
. For example, to test L3C and the baselines, run
python test.py /path/to/logdir 0306_0001,0306_0002,0306_0003 /some/imgdir,/some/other/imgdir \
--names "L3C,RGB Shared,RGB" --recursive=auto
For this, you need to download the models to some directory, in the example this is /path/to/logdir
.
To use the entropy coder and get timings for encoding/decoding, use --write_to_files
(this needs torchac
,
see below. If you did not compile torchac
with CUDA support,
disable CUDA by running CUDA_VISIBLE_DEVICES="" python test.py ...
):
python test.py /path/to/logdir 0306_0001 /some/imgdir --write_to_files=files_out_dir
More flags available with python test.py -h
.
We evaluated our model on 500 images randomly selected from the Open Images validation set, and preprocessed like the training data. To compare, please download Open Images evaluation set here.
The evaluation code automatically split images that are too big into non-overlapping crops. By default,
the threshold is set to images bigger than 2000 * 1500
pixels in total. This can be overwritten by
setting AC_NEEDS_CROP_DIM
from the console, e.g.,
AC_NEEDS_CROP_DIM=2000,2000 python test.py ...
See auto_crop.py
for details.
During the CVPR Poster session, the following where the most frequently asked questions:
First of all, to do lossless compression, you just need to know a probability distribution over your symbols. This is visualized in the bottom left of the poster. Given such a distribution, you can do entropy coding, for example Huffman Coding.
For natural images, we use the pixels of an image as the stream of symbols. Because they are not independent, we model the joint p tilde (x_1, ..., x_N)
(see bottom left of poster).
Now, what Fig. 1 in the paper and the figure in the middle of the poster show is how we learn this p tilde. The important thing to realize is that the output of our model is p tilde, it is not a quantized autoencoder. Given this p tilde, we can do entropy coding. It doesn't matter that we have quantization in the network: no matter how bad p tilde is, you can always do lossless compression -- you might just use a lot of bits!
Note that the model we learn (shown in Fig. 1), is used to get p tilde. We then use this to do entropy coding, as visualized on the top right of the poster or Fig. A4 in the paper.
I'll explain this via Huffman coding, as more people are familiar with that. In the default lossless compression case, we assume the symbols in our stream ("message") are all independent and identically distributed (i.i.d) according to p. We create one Huffman table for all symbols. Consider now the fruit in the bottom left of the poster. Imagine that they are not independent, but that e.g. apples are more likely to appear after bananas. In these cases, it makes sense to have different Huffman tables for different positions in the stream. We would call that "adaptive" Huffman coding. The table at some point in the stream would depend on the conditional distribution of that symbol.
Adaptive Arithmetic coding is the same thing, except that we generalize Arithmetic coding to the non-i.i.d. case.
This is also described in the Paper, in Section 3.1.
Feel free to write an email to us (E-Mail in paper) or open an issue here.
Whenever train.py
is executed, a new experiment is started.
Every experiment is based on a specific configuration file for the network, stored in configs/ms and
another file for the dataloading, stored in configs/dl.
An experiment is uniquely identified by the log date, which is just date and time (e.g. 0506_1107
).
The config files are parsed with the parser from fjcommon
,
which allows hiararchies of configs.
Additionally, there is global_config.py
, to allow quick changes by passing
additional parameters via the -p
flag, which are then available everywhere in the code, see below.
The config files plus the global_config
flags specify all parameters needed to train a network.
When an experiment is started, a directory with all this information is created in the folder passed as
LOG_DIR_ROOT
to train.py
(see python train.py -h
).
For example, running
python train.py configs/ms/cr.cf configs/dl/oi.cf log_dir
results in a folder log_dir
, and in there another folder called
0502_1213 cr oi
If you want to modify the code, consider using the -p
flag. For example,
if you wanted to add a flag for upsampling, you could do
python train.py configs/ms/cr.cf configs/dl/oi.cf log_dir -p upsampling=newfancymethod
which would result in a folder called
0502_1213 cr oi upsampling=newfancymethod
Checkpoints (weights) will be stored in a subfolder called ckpts
.
This experiment can then be evaluated simply by passing the log date to test.py
, in addition to some image folders:
python test.py logs 0502_1213 data/openimages_test,data/raise1k
where we test on images in data/openimages_test
and data/raise1k
.
To use another model as a pretrained model, use --restore
and --restore_restart
:
python train.py configs/ll/cr.cf configs/dl/oi.cf logs --restore 0502_1213 --restore_restart
Name in Paper | Symbol | Name in Code | Short | Class |
---|---|---|---|---|
Feature Extractor | E |
Encoder | enc |
EDSRLikeEnc |
Predictor | D |
Decoder | dec |
EDSRLikeDec |
Quantizer | Q |
Quantizer | q |
Quantizer |
Final box, outputting pi, mu, sigma | Probability Classifier | prob_clf |
AtrousProbabilityClassifier |
See also the notes in src/multiscale_network/multiscale.py
.
The code is quite modular, as it was used to experiment with different things. At the heart is the
MultiscaleBlueprint
class, which has the following main functions: forward
, get_loss
, sample
. It is used by the
MultiscaleTrainer
and MultiscaleTester
. The network is created by MultiscaleNetwork
, which pulls together all
the PyTorch modules needed. The discretized mixture of logistics loss is in DiscretizedMixLogisticsLoss
, which is
usally referred to as dmll
or dmol
in the code.
For bitcoding, there is the Bitcoding
class, which uses the ArithmeticCoding
class, which in turn uses my
torchac
module, written in C++, and described below.
❗ Update: We released torchac
as a stand-alone repo, that does not depend on CUDA.
Check it out here.
To run L3C, you need the version here, but for your own future work, the stand-alone repo will be better. ❗
We implemented an entropy coding module as a C++ extension for PyTorch, because no existing fast Python entropy
coding module was available. You'll need to build it if you plan to use the --write_to_file
flag for test.py
(see Evaluation of Models).
The implementation is based on this blog post,
meaning that we implement arithmetic coding.
It is not optimized, however, it's much faster than doing the equivalent thing in pure-Python (because of all the
bit-shift etc.). Encoding an entire 512 x 512
image happens in 0.202s (see Appendix A in the paper).
The module can be built with or without CUDA. The only difference between the CUDA and non-CUDA versions is:
With CUDA, _get_uint16_cdf
from torchac.py
is done with a simple/non-optimized CUDA kernel (torchac_kernel.cu
),
which has one benefit: we can directly write into shared memory! This saves an expensive copying step from GPU to CPU.
However, compiling with CUDA is probably a hassle. We tested with
- GCC 5.5 and NVCC 9.0
- GCC 7.4 and NVCC 10.1 (update 2)
- Did not work: GCC 6.0 and NVCC 9
Please comment if you have insights into which other configurations work (or don't.)
❗ See update above on the stand-alond torchac repo that does not need GCC or NVCC. ❗
The main part (arithmetic coding), is always on CPU.
Step 1: Make sure a recent gcc
is available in $PATH
by running gcc --version
(tested with version 5.5).
If you want CUDA support, make sure nvcc -V
gives the desired version (tested with nvcc version 9.0).
Step 1b, macOS only (tested with 10.14): Set the following
export CC="clang++ -std=libc++"
export MACOSX_DEPLOYMENT_TARGET=10.14
Step 2:
conda activate l3c_env
cd src/torchac
COMPILE_CUDA=auto python setup.py install
COMPILE_CUDA=auto
: Use CUDA if agcc
between 5 and 6, andnvcc
9 is avaiableCOMPILE_CUDA=force
: Use CUDA, don't checkgcc
ornvcc
COMPILE_CUDA=no
: Don't use CUDA
This installs a package called torchac-backend-cpu
or torchac-backend-gpu
in your pip
.
Both can be installed simultaneously. See also next subsection.
Step 3: To test if it works, you can do
conda activate l3c_env
cd src/torchac
python -c "import torchac"
It should not print any error messages.
Installing torchac-backend-cpu
is easiest. However, if a GPU is available in the system, torchac-backend-gpu
will be faster.
If you use l3c.py
, it will automatically select whether the
code should run on GPU or CPU depending on whether torchac-backend-gpu
is available. The behavior of this can be
tuned with the --device
flag to l3c.py
, e.g., python l3c.py --device=cpu enc ...
, see python l3c.py --help
.
If you use test.py
with the --write_to_files
flag, a check will be performed an exception will be thrown, if the wrong
combination of CUDA available and installed torchac
exists. If you just have torchac-backend-cpu
but a GPU in the system,
disable it via CUDA_VISIBLE_DEVICES="" python test.py ...
.
To sample from L3C, use test.py
with --sample
:
python test.py /path/to/logdir 0306_0001 /some/imgdir --sample=samples
This produces outputs in a directory samples
. Per image, you'll get something like
# Ground Truth
0_IMGNAME_3.549_gt.png
# Sampling from RGB scale, resulting bitcost 1.013bpsp
0_IMGNAME_rgb_1.013.png
# Sampling from RGB scale and z1, resulting bitcost 0.342bpsp
0_IMGNAME_rgb+bn0_0.342.png
# Sampling from RGB scale and z1 and z2, resulting bitcost 0.121bpsp
0_IMGNAME_rgb+bn0+bn1_0.121.png
See Section 5.4. ("Sampling Representations") in the paper.
To encode/decode a single image, use l3c.py
. This requires torchac
:
# Encode to out.l3c
python l3c.py /path/to/logdir 0306_0001 enc /path/to/img out.l3c
# Decode from out.l3c, save to decoded.png
python l3c.py /path/to/logdir 0306_0001 dec out.l3c decoded.png
Use the prep_openimages.sh
script. Run it in an environment with
Python 3,
skimage
(pip install scikit-image
, tested with version 0.13.1), and
awscli
(pip install awscli
):
cd src
./prep_openimages.sh <DATA_DIR>
NOTE: The preprocessing may take a long time. We run it over our internal CPU cluster. Please see
import_train_images.py
for tips on how to incorporate your own cluster.
This will download all images to DATA_DIR
. Make sure there is enough space there, as this script will create
around 300 GB of data. Also, it will probably run for a few hours.
After ./prep_openimages.sh
is done, training data is in DATA_DIR/train_oi
and DATA_DIR/val_oi
. Follow the
instructions printed by ./prep_openimages.sh
to update the config file. You may rm -rf DATA_DIR/download
and
rm -rf DATA_DIR/imported
to free up some space.
(Optional) It helps to have one fixed validation image to monitor training. You may put any image at
- Add support for
nn.DataParallel
. - Incorporate TensorBoard support from PyTorch, instead of pip package.
If you use the work released here for your research, please cite this paper:
@inproceedings{mentzer2019practical,
Author = {Mentzer, Fabian and Agustsson, Eirikur and Tschannen, Michael and Timofte, Radu and Van Gool, Luc},
Booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
Title = {Practical Full Resolution Learned Lossless Image Compression},
Year = {2019}}