Multi-Target Adversarial Frameworks for Domain Adaptation in Semantic Segmentation

Related tags

Deep LearningMTAF
Overview

Multi-Target Adversarial Frameworks for Domain Adaptation in Semantic Segmentation

Paper

Multi-Target Adversarial Frameworks for Domain Adaptation in Semantic Segmentation
Antoine Saporta, Tuan-Hung Vu, Matthieu Cord, Patrick Pérez
valeo.ai, France
IEEE International Conference on Computer Vision (ICCV), 2021 (Poster)

If you find this code useful for your research, please cite our paper:

@inproceedings{saporta2021mtaf,
  title={Multi-Target Adversarial Frameworks for Domain Adaptation in Semantic Segmentation},
  author={Saporta, Antoine and Vu, Tuan-Hung and Cord, Mathieu and P{\'e}rez, Patrick},
  booktitle={ICCV},
  year={2021}
}

Abstract

In this work, we address the task of unsupervised domain adaptation (UDA) for semantic segmentation in presence of multiple target domains: The objective is to train a single model that can handle all these domains at test time. Such a multi-target adaptation is crucial for a variety of scenarios that real-world autonomous systems must handle. It is a challenging setup since one faces not only the domain gap between the labeled source set and the unlabeled target set, but also the distribution shifts existing within the latter among the different target domains. To this end, we introduce two adversarial frameworks: (i) multi-discriminator, which explicitly aligns each target domain to its counterparts, and (ii) multi-target knowledge transfer, which learns a target-agnostic model thanks to a multi-teacher/single-student distillation mechanism.The evaluation is done on four newly-proposed multi-target benchmarks for UDA in semantic segmentation. In all tested scenarios, our approaches consistently outperform baselines, setting competitive standards for the novel task.

Preparation

Pre-requisites

  • Python 3.7
  • Pytorch >= 0.4.1
  • CUDA 9.0 or higher

Installation

  1. Clone the repo:
$ git clone https://github.com/valeoai/MTAF
$ cd MTAF
  1. Install OpenCV if you don't already have it:
$ conda install -c menpo opencv
  1. Install NVIDIA Apex if you don't already have it: follow the instructions on: https://github.com/NVIDIA/apex

  2. Install this repository and the dependencies using pip:

$ pip install -e <root_dir>

With this, you can edit the MTAF code on the fly and import function and classes of MTAF in other project as well.

  1. Optional. To uninstall this package, run:
$ pip uninstall MTAF

Datasets

By default, the datasets are put in <root_dir>/data. We use symlinks to hook the MTAF codebase to the datasets. An alternative option is to explicitlly specify the parameters DATA_DIRECTORY_SOURCE and DATA_DIRECTORY_TARGET in YML configuration files.

  • GTA5: Please follow the instructions here to download images and semantic segmentation annotations. The GTA5 dataset directory should have this basic structure:
<root_dir>/data/GTA5/                               % GTA dataset root
<root_dir>/data/GTA5/images/                        % GTA images
<root_dir>/data/GTA5/labels/                        % Semantic segmentation labels
...
  • Cityscapes: Please follow the instructions in Cityscape to download the images and ground-truths. The Cityscapes dataset directory should have this basic structure:
<root_dir>/data/cityscapes/                         % Cityscapes dataset root
<root_dir>/data/cityscapes/leftImg8bit              % Cityscapes images
<root_dir>/data/cityscapes/leftImg8bit/train
<root_dir>/data/cityscapes/leftImg8bit/val
<root_dir>/data/cityscapes/gtFine                   % Semantic segmentation labels
<root_dir>/data/cityscapes/gtFine/train
<root_dir>/data/cityscapes/gtFine/val
...
  • Mapillary: Please follow the instructions in Mapillary Vistas to download the images and validation ground-truths. The Mapillary Vistas dataset directory should have this basic structure:
<root_dir>/data/mapillary/                          % Mapillary dataset root
<root_dir>/data/mapillary/train                     % Mapillary train set
<root_dir>/data/mapillary/train/images
<root_dir>/data/mapillary/validation                % Mapillary validation set
<root_dir>/data/mapillary/validation/images
<root_dir>/data/mapillary/validation/labels
...
  • IDD: Please follow the instructions in IDD to download the images and validation ground-truths. The IDD Segmentation dataset directory should have this basic structure:
<root_dir>/data/IDD/                         % IDD dataset root
<root_dir>/data/IDD/leftImg8bit              % IDD images
<root_dir>/data/IDD/leftImg8bit/train
<root_dir>/data/IDD/leftImg8bit/val
<root_dir>/data/IDD/gtFine                   % Semantic segmentation labels
<root_dir>/data/IDD/gtFine/val
...

Pre-trained models

Pre-trained models can be downloaded here and put in <root_dir>/pretrained_models

Running the code

For evaluation, execute:

$ cd <root_dir>/mtaf/scripts
$ python test.py --cfg ./configs/gta2cityscapes_mapillary_baseline_pretrained.yml
$ python test.py --cfg ./configs/gta2cityscapes_mapillary_mdis_pretrained.yml
$ python test.py --cfg ./configs/gta2cityscapes_mapillary_mtkt_pretrained.yml

Training

For the experiments done in the paper, we used pytorch 1.3.1 and CUDA 10.0. To ensure reproduction, the random seed has been fixed in the code. Still, you may need to train a few times to reach the comparable performance.

By default, logs and snapshots are stored in <root_dir>/experiments with this structure:

<root_dir>/experiments/logs
<root_dir>/experiments/snapshots

To train the multi-target baseline:

$ cd <root_dir>/mtaf/scripts
$ python train.py --cfg ./configs/gta2cityscapes_mapillary_baseline.yml

To train the Multi-Discriminator framework:

$ cd <root_dir>/mtaf/scripts
$ python train.py --cfg ./configs/gta2cityscapes_mapillary_mdis.yml

To train the Multi-Target Knowledge Transfer framework:

$ cd <root_dir>/mtaf/scripts
$ python train.py --cfg ./configs/gta2cityscapes_mapillary_mtkt.yml

Testing

To test the multi-target baseline:

$ cd <root_dir>/mtaf/scripts
$ python test.py --cfg ./configs/gta2cityscapes_mapillary_baseline.yml

To test the Multi-Discriminator framework:

$ cd <root_dir>/mtaf/scripts
$ python test.py --cfg ./configs/gta2cityscapes_mapillary_mdis.yml

To test the Multi-Target Knowledge Transfer framework:

$ cd <root_dir>/mtaf/scripts
$ python test.py --cfg ./configs/gta2cityscapes_mapillary_mtkt.yml

Acknowledgements

This codebase is heavily borrowed from ADVENT.

License

MTAF is released under the Apache 2.0 license.

Comments
  • question about adversarial training code in train_UDA.py

    question about adversarial training code in train_UDA.py

    Thank you for sharing the code for your excellent work. I have some basic questions about your implementation. pred_trg_main = interp_target(all_pred_trg_main[i+1]) ## what does [i+1] mean? pred_trg_main_list.append(pred_trg_main) pred_trg_target = interp_target(all_pred_trg_main[0]) ## what does [0] mean? pred_trg_target_list.append(pred_trg_target)

    In train_UDA.py, line 829-836, why should we use index[i+1] and [0]? What's the meaning of that? Also, where is the definition of the target-agnostic classifier in your code?

    Thanks again and look forward to hearing back from you!

    opened by yuzhang03 2
  • the problem for training loss

    the problem for training loss

    Thanks for enlightening work agian.

    I train the Mdis method for one source and one target, but I am confused for the loss, and I plot by tensorboard. And as I think, the adv loss should walk low and the discrimitor loss should walk higher. but in the loss below, the two losses oscillate around a number. whats wrong with it?

    Besides, I infer the training results should be better when training in manner of 1source 1target instead of 1source multi target. But in my training, I dont get good results.

    So hope your thought sincerely.

    And my training config: adv loss weight: 0.5 adv learning rate: 1e-5 seg learning rate: 1.25e-5

    adversarial loss of one source and one target
    image

    dicriminator loss of one source and one target image

    opened by slz929 2
  • problem for training data

    problem for training data

    Thanks for enlightening and practical work about multi-target DA ! I have read your paper, and I found one source dataset and 3 target datasets of unequal quantity, does the quantity of data for every domain matters? And what is the appropriate amount of training data for MTKT? Another question, I want to know why KL loss is used for knowledge transfer? If I want to train an embedding word instead of a segmentation map, is the KL loss appropriate, and is there a better alternative?

    opened by slz929 2
  • About the generation of segmentation color maps

    About the generation of segmentation color maps

    Thanks for the great research!

    I have a question though, the mIoU you report in your paper is for 7 classes, but the segmentation colour map in the qualitative analysis seems to be for the 19 classes commonly used in domain adaptive semantic segmentation.

    In other words, how can a model trained on 7 classes be used to generate a 19-class segmentation colour map? Or am I wrong in my understanding?

    I look forward to your response.

    Thank you!

    opened by liwei1101 1
  • About labels of IDD dataset

    About labels of IDD dataset

    Hello! @SportaXD Thank you for your great work!

    I was reproducing the code and noticed: the labels in the IDD dataset are in JSON file format instead of segmentation label form.

    How is this problem solved?

    opened by liwei1101 1
  • About MTKT code

    About MTKT code

    In train_UDA.py 758 line

            d_main_list[i] = d_main
            optimizer_d_main_list.append(optimizer_d_main)
            d_aux_list[i] = d_aux
            optimizer_d_aux_list.append(optimizer_d_aux)
    

    If this were done(d_main_list[i] = d_main and d_aux_list[i] = d_aux), it would make all the discriminators in the list use the same one, shouldn't there be one discriminator for each classifier?

    opened by liwei1101 1
  • About 'the multi-target baseline'

    About 'the multi-target baseline'

    Thank you for sharing the code for your excellent work. I have some basic questions about your implementation.

    d_main = get_fc_discriminator(num_classes=num_classes)
    d_main.train()
    d_main.to(device)
    d_aux = get_fc_discriminator(num_classes=num_classes)
    d_aux.train()
    d_aux.to(device)
    

    Can you tell me why the multi-domain baseline code does not use multiple discriminators but only one discriminator. It looks like a single domain approach. Thanks!

    opened by liwei1101 1
  • about eval_UDA.py

    about eval_UDA.py

    Thanks for sharing your codes.

    I was impressed with your good research.

    Could you explain why the output map is not resized for target size(cfg.TEST.OUTPUT_SIZE_TARGET) in the case of Mapillary dataset in line 57 of eval_UDA.py?

    When I tested the trained model on Mapillary dataset, inference took a long time due to the large resolution.

    I'm looking forward to hearing from you.

    Thank you!

    opened by jdg900 1
  • modifying info7class.json and train_UDA.py

    modifying info7class.json and train_UDA.py

    we have found a small bug in "./MTAF/mtaf/dataset/cityscapes_list/info7class.json". valeo

    It should be 7 Classes rather than 19 Classes in the configuration file. It appears in the Evaluation stage, where the result is printed out in the mIoU evaluation metrics and the names of the 7 classes.

    Also, there is a typo in the comments.

    opened by mohamedelmesawy 1
  • Running MTAF on a slightly different setup

    Running MTAF on a slightly different setup

    Hello, thanks for sharing the code and such a good contribution. I would like to run your method on a setup that is a bit different, specifically adapting from Cityscapes ---> BDD, Mapillary. I have seen that the code accepts Cityscapes for both source and target, so that shouldnt be a problem, and I have added a dataloader for BDD to be the target 1.

    In order to get the best performance, do I need to train the baseline and then train the method using MTKT or MDIS loading the baseline as pretrained? Or do I get the best performance directly by running the training script for MTKT or MDIS without the baseline?

    opened by fabriziojpiva 1
Owner
Valeo.ai
The GitHub account of Valeo.ai
Valeo.ai
Reproduction of Vision Transformer in Tensorflow2. Train from scratch and Finetune.

Vision Transformer(ViT) in Tensorflow2 Tensorflow2 implementation of the Vision Transformer(ViT). This repository is for An image is worth 16x16 words

sungjun lee 42 Dec 27, 2022
Jupyter notebooks for using & learning Keras

deep-learning-with-keras-notebooks 這個github的repository主要是個人在學習Keras的一些記錄及練習。希望在學習過程中發現到一些好的資訊與範例也可以對想要學習使用 Keras來解決問題的同好,或是對深度學習有興趣的在學學生可以有一些方便理解與上手範例

ErhWen Kuo 2.1k Dec 27, 2022
Global-Local Context Network for Person Search

Global-Local Context Network for Person Search Abstract: Person search aims to jointly localize and identify a query person from natural, uncropped im

Peng Zheng 15 Oct 17, 2022
Tool for installing and updating MiSTer cores and other files

MiSTer Downloader This tool installs and updates all the cores and other extra files for your MiSTer. It also updates the menu core, the MiSTer firmwa

72 Dec 24, 2022
Deep Sea Treasure Environment for Multi-Objective Optimization Research

DeepSeaTreasure Environment Installation In order to get started with this environment, you can install it using the following command: python3 -m pip

imec IDLab 6 Nov 14, 2022
Simple implementation of OpenAI CLIP model in PyTorch.

It was in January of 2021 that OpenAI announced two new models: DALL-E and CLIP, both multi-modality models connecting texts and images in some way. In this article we are going to implement CLIP mod

Moein Shariatnia 226 Jan 05, 2023
High level network definitions with pre-trained weights in TensorFlow

TensorNets High level network definitions with pre-trained weights in TensorFlow (tested with 2.1.0 = TF = 1.4.0). Guiding principles Applicability.

Taehoon Lee 1k Dec 13, 2022
Create UIs for prototyping your machine learning model in 3 minutes

Note: We just launched Hosted, where anyone can upload their interface for permanent hosting. Check it out! Welcome to Gradio Quickly create customiza

Gradio 11.7k Jan 07, 2023
Local Multi-Head Channel Self-Attention for FER2013

LHC-Net Local Multi-Head Channel Self-Attention This repository is intended to provide a quick implementation of the LHC-Net and to replicate the resu

12 Jan 04, 2023
Contains code for the paper "Vision Transformers are Robust Learners".

Vision Transformers are Robust Learners This repository contains the code for the paper Vision Transformers are Robust Learners by Sayak Paul* and Pin

Sayak Paul 103 Jan 05, 2023
PyTorch implementation for the paper Pseudo Numerical Methods for Diffusion Models on Manifolds

Pseudo Numerical Methods for Diffusion Models on Manifolds (PNDM) This repo is the official PyTorch implementation for the paper Pseudo Numerical Meth

Luping Liu (刘路平) 196 Jan 05, 2023
Mercury: easily convert Python notebook to web app and share with others

Mercury Share your Python notebooks with others Easily convert your Python notebooks into interactive web apps by adding parameters in YAML. Simply ad

MLJAR 2.2k Dec 27, 2022
Implementation for paper MLP-Mixer: An all-MLP Architecture for Vision

MLP Mixer Implementation for paper MLP-Mixer: An all-MLP Architecture for Vision. Give us a star if you like this repo. Author: Github: bangoc123 Emai

Ngoc Nguyen Ba 86 Dec 10, 2022
Finite-temperature variational Monte Carlo calculation of uniform electron gas using neural canonical transformation.

CoulombGas This code implements the neural canonical transformation approach to the thermodynamic properties of uniform electron gas. Building on JAX,

FermiFlow 9 Mar 03, 2022
PyTorch implementation of our paper: Decoupling and Recoupling Spatiotemporal Representation for RGB-D-based Motion Recognition

Decoupling and Recoupling Spatiotemporal Representation for RGB-D-based Motion Recognition, arxiv This is a PyTorch implementation of our paper. 1. Re

DamoCV 11 Nov 19, 2022
code for paper -- "Seamless Satellite-image Synthesis"

Seamless Satellite-image Synthesis by Jialin Zhu and Tom Kelly. Project site. The code of our models borrows heavily from the BicycleGAN repository an

Light 14 Apr 05, 2022
Data-Driven Operational Space Control for Adaptive and Robust Robot Manipulation

OSCAR Project Page | Paper This repository contains the codebase used in OSCAR: Data-Driven Operational Space Control for Adaptive and Robust Robot Ma

NVIDIA Research Projects 74 Dec 22, 2022
Learning Chinese Character style with conditional GAN

zi2zi: Master Chinese Calligraphy with Conditional Adversarial Networks Introduction Learning eastern asian language typefaces with GAN. zi2zi(字到字, me

Yuchen Tian 2.2k Jan 02, 2023
LoveDA: A Remote Sensing Land-Cover Dataset for Domain Adaptive Semantic Segmentation

LoveDA: A Remote Sensing Land-Cover Dataset for Domain Adaptive Semantic Segmentation by Junjue Wang, Zhuo Zheng, Ailong Ma, Xiaoyan Lu, and Yanfei Zh

Payphone 8 Nov 21, 2022
Implementation of Neural Style Transfer in Pytorch

PytorchNeuralStyleTransfer Code to run Neural Style Transfer from our paper Image Style Transfer Using Convolutional Neural Networks. Also includes co

Leon Gatys 396 Dec 01, 2022