Reference PyTorch implementation of "End-to-end optimized image compression with competition of prior distributions"

Related tags

Deep Learningneural-networksimage-compressioncdf
Overview

PyTorch reference implementation of "End-to-end optimized image compression with competition of prior distributions" by Benoit Brummer and Christophe De Vleeschouwer ( https://github.com/trougnouf/Manypriors )

Forked from PyTorch implementation of "Variational image compression with a scale hyperprior" by Jiaheng Liu ( https://github.com/liujiaheng/compression )

This code is experimental.

Requirements

TODO torchac should be switched to the standalone release on https://github.com/fab-jul/torchac (which was not yet released at the time of writing this code)

Arch

pacaur -S python-tqdm python-pytorch-torchac python-configargparse python-yaml python-ptflops python-colorspacious python-pypng python-pytorch-piqa-git

Ubuntu / Slurm cluster / misc:

TMPDIR=tmp pip3 install --user torch==1.7.0+cu92 torchvision==0.8.1+cu92 -f https://download.pytorch.org/whl/torch_stable.html
TMPDIR=tmp pip3 install --user tqdm matplotlib tensorboardX scipy scikit-image scikit-video ConfigArgParse pyyaml h5py ptflops colorspacious pypng piqa

torchac must be compiled and installed per https://github.com/trougnouf/L3C-PyTorch/tree/master/src/torchac

torchac $ COMPILE_CUDA=auto python3 setup.py build
torchac $ python3 setup.py install --optimize=1 --skip-build

or (untested)

torchac $ pip install .

Once Ubuntu updates PyTorch then tensorboardX won't be required

Dataset gathering

Copy the kodak dataset into datasets/test/kodak

cd ../common
python tools/wikidownloader.py --category "Category:Featured pictures on Wikimedia Commons"
python tools/wikidownloader.py --category "Category:Formerly featured pictures on Wikimedia Commons"
python tools/wikidownloader.py --category "Category:Photographs taken on Ektachrome and Elite Chrome film"
mv "../../datasets/Category:Featured pictures on Wikimedia Commons" ../../datasets/FeaturedPictures
mv "../../datasets/Category:Formerly featured pictures on Wikimedia Commons" ../../datasets/Formerly_featured_pictures_on_Wikimedia_Commons
mv "../../datasets/Category:Photographs taken on Ektachrome and Elite Chrome film" ../../datasets/Photographs_taken_on_Ektachrome_and_Elite_Chrome_film
python tools/verify_images.py ../../datasets/FeaturedPictures/
python tools/verify_images.py ../../datasets/Formerly_featured_pictures_on_Wikimedia_Commons/
python tools/verify_images.py ../../datasets/Photographs_taken_on_Ektachrome_and_Elite_Chrome_film/

# TODO make a list of train/test img automatically s.t. images don't have to be copied over the network

Crop images to 1024*1024. from src/common: (in python)

import os
from libs import libdsops
for ads in ['Formerly_featured_pictures_on_Wikimedia_Commons', 'Photographs_taken_on_Ektachrome_and_Elite_Chrome_film', 'FeaturedPictures']:
    libdsops.split_traintest(ads)
    libdsops.crop_ds_dpath(ads, 1024, root_ds_dpath=os.path.join(libdsops.ROOT_DS_DPATH, 'train'), num_threads=os.cpu_count()//2)

#verify crops
python3 tools/verify_images.py ../../datasets/train/resized/1024/FeaturedPictures/
python3 tools/verify_images.py ../../datasets/train/resized/1024/Formerly_featured_pictures_on_Wikimedia_Commons/
python3 tools/verify_images.py ../../datasets/train/resized/1024/Photographs_taken_on_Ektachrome_and_Elite_Chrome_film/
# use the --save_img flag at the end of verify_images.py commands if training fails after the simple verification

Move a small subset of the training cropped images to a matching test directory and use it as args.val_dpath

JPEG/BPG compression of the Commons Test Images is done with common/tools/bpg_jpeg_compress_commons.py and comp/tools/bpg_jpeg_test_commons.py

Loading

Loading a model: provide all necessary (non-default) parameters s.a. arch, num_distributions, etc. Saved yaml can be used iff the ConfigArgParse patch from https://github.com/trougnouf/ConfigArgParse is applied, otherwise unset values are overwritten with the "None" string.

Training

Train a base model (given arch and num_distributions) for 6M steps at train_lambda=4096, fine-tune for 4M steps with lower train_lambda and/or msssim lossf Set arch to Manypriors for this work, use num_distributions 1 for Balle2017, or set arch to Balle2018PTTFExp for Balle2018 (hyperprior) egrun:

python train.py --num_distributions 64 --arch ManyPriors --train_lambda 4096 --expname mse_4096_manypriors_64_CLI
# and/or
python train.py --config configs/mse_4096_manypriors_64pr.yaml
# and/or
python train.py --config configs/mse_2048_manypriors_64pr.yaml --pretrain mse_4096_manypriors_64pr --reset_lr --reset_global_step # --reset_optimizer
# and/or
python train.py --config configs/mse_4096_hyperprior.yaml

--passthrough_ae is now activated by default. It was not used in the paper, but should result in better rate-distortion. To turn it off, change config/defaults.yaml or use --no_passthrough_ae

Tests

egruns: Test complexity:

python tests.py --complexity --pretrain mse_4096_manypriors_64pr --arch ManyPriors --num_distributions 64

Test timing:

python tests.py --timing "../../datasets/test/Commons_Test_Photographs" --pretrain mse_4096_manypriors_64pr --arch ManyPriors --num_distributions 64

Segment the images in commons_test_dpath by distribution index:

python tests.py --segmentation --commons_test_dpath "../../datasets/test/Commons_Test_Photographs" --pretrain mse_4096_manypriors_64pr --arch ManyPriors --num_distributions 64

Visualize cumulative distribution functions:

python tests.py --plot --pretrain mse_4096_manypriors_64pr --arch ManyPriors --num_distributions 64

Test on kodak images:

python tests.py --encdec_kodak --test_dpath "../../datasets/test/kodak/" --pretrain mse_4096_manypriors_64pr --arch ManyPriors --num_distributions 64

Test on commons images (larger, uses CPU):

python tests.py --encdec_commons --test_commons_dpath "../../datasets/test/Commons_Test_Photographs/" --pretrain checkpoints/mse_4096_manypriors_64pr/saved_models/checkpoint.pth --arch ManyPriors --num_distributions 64

Encode an image:

python tests.py --encode "../../datasets/test/Commons_Test_Photographs/Garden_snail_moving_down_the_Vennbahn_in_disputed_territory_(DSCF5879).png" --pretrain mse_4096_manypriors_64pr --arch ManyPriors --num_distributions 64 --device -1

Decode that image:

python tests.py --decode "checkpoints/mse_4096_manypriors_64pr/encoded/Garden_snail_moving_down_the_Vennbahn_in_disputed_territory_(DSCF5879).png" --pretrain mse_4096_manypriors_64pr --arch ManyPriors --num_distributions 64 --device -1
Owner
Benoit Brummer
BS CpE at @UCF (2016), MS CS (AI) @uclouvain (2019), PhD student @uclouvain w/ intoPIX
Benoit Brummer
GitHub Repository
Tensorflow 2 Object Detection API kurulumu, GPU desteği, custom model hazırlama

Tensorflow 2 Object Detection API Bu tutorial, TensorFlow 2.x'in kararlı sürümü olan TensorFlow 2.3'ye yöneliktir. Bu, görüntülerde / videoda nesne a

46 Nov 20, 2022
System Combination for Grammatical Error Correction Based on Integer Programming

System Combination for Grammatical Error Correction Based on Integer Programming This repository contains the code and scripts that implement the syst

NUS NLP Group 0 Mar 29, 2022
Unsupervised Pre-training for Person Re-identification (LUPerson)

LUPerson Unsupervised Pre-training for Person Re-identification (LUPerson). The repository is for our CVPR2021 paper Unsupervised Pre-training for Per

143 Dec 24, 2022
Meta-learning for NLP

Self-Supervised Meta-Learning for Few-Shot Natural Language Classification Tasks Code for training the meta-learning models and fine-tuning on downstr

IESL 43 Nov 08, 2022
Finite difference solution of 2D Poisson equation. Can handle Dirichlet, Neumann and mixed boundary conditions.

Poisson-solver-2D Finite difference solution of 2D Poisson equation Current version can handle Dirichlet, Neumann, and mixed (combination of Dirichlet

Mohammad Asif Zaman 34 Dec 23, 2022
[ICCV'21] Pri3D: Can 3D Priors Help 2D Representation Learning?

Pri3D: Can 3D Priors Help 2D Representation Learning? [ICCV 2021] Pri3D leverages 3D priors for downstream 2D image understanding tasks: during pre-tr

Ji Hou 124 Jan 06, 2023
Code for the Higgs Boson Machine Learning Challenge organised by CERN & EPFL

A method to solve the Higgs boson challenge using Least Squares - Novae This project is the Project 1 of EPFL CS-433 Machine Learning. The project is

Giacomo Orsi 1 Nov 09, 2021
Repository sharing code and the model for the paper "Rescoring Sequence-to-Sequence Models for Text Line Recognition with CTC-Prefixes"

Rescoring Sequence-to-Sequence Models for Text Line Recognition with CTC-Prefixes Setup virtualenv -p python3 venv source venv/bin/activate pip instal

Planet AI GmbH 9 May 20, 2022
A minimal solution to hand motion capture from a single color camera at over 100fps. Easy to use, plug to run.

Minimal Hand A minimal solution to hand motion capture from a single color camera at over 100fps. Easy to use, plug to run. This project provides the

Yuxiao Zhou 824 Jan 07, 2023
《Truly shift-invariant convolutional neural networks》(2021)

Truly shift-invariant convolutional neural networks [Paper] Authors: Anadi Chaman and Ivan Dokmanić Convolutional neural networks were always assumed

Anadi Chaman 46 Dec 19, 2022
Nested cross-validation is necessary to avoid biased model performance in embedded feature selection in high-dimensional data with tiny sample sizes

Pruner for nested cross-validation - Sphinx-Doc Nested cross-validation is necessary to avoid biased model performance in embedded feature selection i

1 Dec 15, 2021
Variational autoencoder for anime face reconstruction

VAE animeface Variational autoencoder for anime face reconstruction Introduction This repository is an exploratory example to train a variational auto

Minzhe Zhang 2 Dec 11, 2021
FLAVR is a fast, flow-free frame interpolation method capable of single shot multi-frame prediction

FLAVR is a fast, flow-free frame interpolation method capable of single shot multi-frame prediction. It uses a customized encoder decoder architecture with spatio-temporal convolutions and channel ga

Tarun K 280 Dec 23, 2022
Breaking the Dilemma of Medical Image-to-image Translation

Breaking the Dilemma of Medical Image-to-image Translation Supervised Pix2Pix and unsupervised Cycle-consistency are two modes that dominate the field

Kid Liet 86 Dec 21, 2022
Enhancing Column Generation by a Machine-Learning-BasedPricing Heuristic for Graph Coloring

Enhancing Column Generation by a Machine-Learning-BasedPricing Heuristic for Graph Coloring (to appear at AAAI 2022) We propose a machine-learning-bas

YunzhuangS 2 May 02, 2022
The official implementation of the CVPR2021 paper: Decoupled Dynamic Filter Networks

Decoupled Dynamic Filter Networks This repo is the official implementation of CVPR2021 paper: "Decoupled Dynamic Filter Networks". Introduction DDF is

F.S.Fire 180 Dec 30, 2022
A web application that provides real time temperature and humidity readings of a house.

About A web application which provides real time temperature and humidity readings of a house. If you're interested in the data collected so far click

Ben Thompson 3 Jan 28, 2022
Code release for "Transferable Semantic Augmentation for Domain Adaptation" (CVPR 2021)

Transferable Semantic Augmentation for Domain Adaptation Code release for "Transferable Semantic Augmentation for Domain Adaptation" (CVPR 2021) Paper

66 Dec 16, 2022
A Comparative Framework for Multimodal Recommender Systems

Cornac Cornac is a comparative framework for multimodal recommender systems. It focuses on making it convenient to work with models leveraging auxilia

Preferred.AI 671 Jan 03, 2023
Official implement of Evo-ViT: Slow-Fast Token Evolution for Dynamic Vision Transformer

Evo-ViT: Slow-Fast Token Evolution for Dynamic Vision Transformer This repository contains the PyTorch code for Evo-ViT. This work proposes a slow-fas

YifanXu 53 Dec 05, 2022