PyTorch implementation of: Michieli U. and Zanuttigh P., "Continual Semantic Segmentation via Repulsion-Attraction of Sparse and Disentangled Latent Representations", CVPR 2021.

Related tags

Deep LearningSDR
Overview

Continual Semantic Segmentation via Repulsion-Attraction of Sparse and Disentangled Latent Representations

This is the official PyTorch implementation of our work: "Continual Semantic Segmentation via Repulsion-Attraction of Sparse and Disentangled Latent Representations" published at CVPR 2021.

In this paper, we present some novel approaches constraining the feature space for continual learning semantic segmentation models. The evaluation on Pascal VOC2012 and on ADE20K validated our method.

Paper
5-min video
slides
poster
teaser

Requirements

This repository uses the following libraries:

  • Python (3.7.6)
  • Pytorch (1.4.0) [tested up to 1.7.1]
  • torchvision (0.5.0)
  • tensorboardX (2.0)
  • matplotlib (3.1.1)
  • numpy (1.18.1)
  • apex (0.1) [optional]
  • inplace-abn (1.0.7) [optional]

We also assume to have installed pytorch.distributed package.

All the dependencies are listed in the requirements.txt file which can be used in conda as:
conda create --name <env> --file requirements.txt

How to download data

In this project we use two dataset, ADE20K and Pascal-VOC 2012. We provide the scripts to download them in 'data/download_<dataset_name>.sh'. The script takes no inputs and you should use it in the target directory (where you want to download data).

How to perform training

The most important file is run.py, that is in charge to start the training or test procedure. To run it, simpy use the following command:

python -m torch.distributed.launch --nproc_per_node=<num_GPUs> run.py --data_root <data_folder> --name <exp_name> .. other args ..

The default is to use a pretraining for the backbone used, which is the one officially released by PyTorch models and it will be downloaded automatically. If you don't want to use pretrained, please use --no-pretrained.

There are many options and you can see them all by using --help option. Some of them are discussed in the following:

  • please specify the data folder using: --data_root <data_root>
  • dataset: --dataset voc (Pascal-VOC 2012) | ade (ADE20K)
  • task: --task <task>, where tasks are
    • 15-5, 15-5s, 19-1 (VOC), 100-50, 100-10, 50, 100-50b, 100-10b, 50b (ADE, b indicates the order)
  • step (each step is run separately): --step <N>, where N is the step number, starting from 0
  • (only for Pascal-VOC) disjoint is default setup, to enable overlapped: --overlapped
  • learning rate: --lr 0.01 (for step 0) | 0.001 (for step > 0)
  • batch size: --batch_size 8 (Pascal-VOC 2012) | 4 (ADE20K)
  • epochs: --epochs 30 (Pascal-VOC 2012) | 60 (ADE20K)
  • method: --method <method name>, where names are
    • FT, LWF, LWF-MC, ILT, EWC, RW, PI, MIB, CIL, SDR
      Note that method overwrites other parameters, but can be used as a kickstart to use default parameters for each method (see more on this in the hyperparameters section below)

For all the details please follow the information provided using the help option.

Example training commands

We provide some example scripts in the *.slurm and *.bat files.
For instance, to run the step 0 of 19-1 VOC2012 you can run:

python -u -m torch.distributed.launch 1> 'outputs/19-1/output_19-1_step0.txt' 2>&1 \
--nproc_per_node=1 run.py \
--batch_size 8 \
--logdir logs/19-1/ \
--dataset voc \
--name FT \
--task 19-1 \
--step 0 \
--lr 0.001 \
--epochs 30 \
--debug \
--sample_num 10 \
--unce \
--loss_de_prototypes 1 \
--where_to_sim GPU_windows

Note: loss_de_prototypes is set to 1 only for having the prototypes computed in the 0-th step (no distillation is actually computed of course).

Then, the step 1 of the same scenario can be computed simply as:

python -u -m torch.distributed.launch 1> 'outputs/19-1/output_19-1_step1.txt'  2>&1 \
--nproc_per_node=1 run.py \
--batch_size 8 \
--logdir logs/19-1/ \
--dataset voc \
--task 19-1 \
--step 1 \
--lr 0.0001 \
--epochs 30 \
--debug \
--sample_num 10 \
--where_to_sim GPU_windows \
--method SDR \
--step_ckpt 'logs/19-1/19-1-voc_FT/19-1-voc_FT_0.pth'

The results obtained are reported inside the outputs/ and logs/ folder, which can be downloaded here, and are 0.4% of mIoU higher than those reported in the main paper due to a slightly changed hyperparameter.

To run other approaches it is sufficient to change the --method parameter into one of the following: FT, LWF, LWF-MC, ILT, EWC, RW, PI, MIB, CIL, SDR.

Note: for the best results, the hyperparameters may change. Please see further details on the hyperparameters section below.

Once you trained the model, you can see the result on tensorboard (we perform the test after the whole training) or on the output files. or you can test it by using the same script and parameters but using the option

--test

that will skip all the training procedure and test the model on test data.

Do you want to try our constraints on your codebase or task?

If you want to try our novel constraints on your codebase or on a different problem you can check the utils/loss.py file. Here, you can take the definitions of the different losses and embed them into your codebase
The names of the variables could be interpreted as:

  • targets-- ground truth map,
  • outputs-- segmentation map output from the current network
  • outputs_old-- segmentation map output from the previous network
  • features-- features taken from the end of the currently-trained encoder,
  • features_old-- features taken from the end of the previous encoder [used for distillation on the encoder on ILT, but not used on SDR],
  • prototypes-- prototypical feature representations
  • incremental_step -- index of the current incremental step (0 if first non-incremental training is performed)
  • classes_old-- index of previous classes

Range for the Hyper-parameters

For what concerns the hyperparameters of our approach:

  • The parameter for the distillation loss is in the same range of that of MiB,
  • Prototypes matching: lambda was searched in range 1e-1 to 1e-3,
  • Contrastive learning (or clustering): lambda was searched in the range of 1e-2 to 1e-3,
  • Features sparsification: lambda was searched in the range of 1e-3 to 1e-5 A kick-start could be to use KD 10, PM 1e-2, CL 1e-3 and FS 1e-4.
    The best parameters may vary across datasets and incremental setup. However, we typically did a grid search and kept it fixed across learning steps.

So, writing explicitly all the parameters, the command would look something like the following:

python -u -m torch.distributed.launch 1> 'outputs/19-1/output_19-1_step1_custom.txt'  2>&1 \
--nproc_per_node=1 run.py \
--batch_size 8 \
--logdir logs/19-1/ \
--dataset voc \
--task 19-1 \
--step 1 \
--lr 0.0001 \
--epochs 30 \
--debug \
--sample_num 10 \
--where_to_sim GPU_windows \
--unce \
--loss_featspars $loss_featspars \
--lfs_normalization $lfs_normalization \
--lfs_shrinkingfn $lfs_shrinkingfn \
--lfs_loss_fn_touse $lfs_loss_fn_touse \
--loss_de_prototypes $loss_de_prototypes \
--loss_de_prototypes_sumafter \
--lfc_sep_clust $lfc_sep_clust \
--loss_fc $loss_fc \
--loss_kd $loss_kd \
--step_ckpt 'logs/19-1/19-1-voc_FT/19-1-voc_FT_0.pth'

Cite us

If you use this repository, please consider to cite

   @inProceedings{michieli2021continual,
   author = {Michieli, Umberto and Zanuttigh, Pietro},
   title  = {Continual Semantic Segmentation via Repulsion-Attraction of Sparse and Disentangled Latent Representations},
   booktitle = {Computer Vision and Pattern Recognition (CVPR)},
   year      = {2021},
   month     = {June}
   }

And our previous works ILT and its journal extension.

Acknowledgements

We gratefully acknowledge the authors of MiB paper for the insightful discussion and for providing the open source codebase, which has been the starting point for our work.
We also acknowledge the authors of CIL for providing their code even before the official release.

Owner
Multimedia Technology and Telecommunication Lab
Department of Information Engineering, University of Padova
Multimedia Technology and Telecommunication Lab
Code-free deep segmentation for computational pathology

NoCodeSeg: Deep segmentation made easy! This is the official repository for the manuscript "Code-free development and deployment of deep segmentation

André Pedersen 26 Nov 23, 2022
Reimplementation of Dynamic Multi-scale filters for Semantic Segmentation.

Paddle implementation of Dynamic Multi-scale filters for Semantic Segmentation.

Hongqiang.Wang 2 Nov 01, 2021
ONNX Runtime: cross-platform, high performance ML inferencing and training accelerator

ONNX Runtime is a cross-platform inference and training machine-learning accelerator. ONNX Runtime inference can enable faster customer experiences an

Microsoft 8k Jan 04, 2023
Decorators for maximizing memory utilization with PyTorch & CUDA

torch-max-mem This package provides decorators for memory utilization maximization with PyTorch and CUDA by starting with a maximum parameter size and

Max Berrendorf 10 May 02, 2022
fcn by tensorflow

Update An example on how to integrate this code into your own semantic segmentation pipeline can be found in my KittiSeg project repository. tensorflo

9 May 22, 2022
QRec: A Python Framework for quick implementation of recommender systems (TensorFlow Based)

Introduction QRec is a Python framework for recommender systems (Supported by Python 3.7.4 and Tensorflow 1.14+) in which a number of influential and

Yu 1.4k Dec 30, 2022
Point Cloud Denoising input segmentation output raw point-cloud valid/clear fog rain de-noised Abstract Lidar sensors are frequently used in environme

Point Cloud Denoising input segmentation output raw point-cloud valid/clear fog rain de-noised Abstract Lidar sensors are frequently used in environme

75 Nov 24, 2022
Multi-modal Content Creation Model Training Infrastructure including the FACT model (AI Choreographer) implementation.

AI Choreographer: Music Conditioned 3D Dance Generation with AIST++ [ICCV-2021]. Overview This package contains the model implementation and training

Google Research 365 Dec 30, 2022
CAST: Character labeling in Animation using Self-supervision by Tracking

CAST: Character labeling in Animation using Self-supervision by Tracking (Published as a conference paper at EuroGraphics 2022) Note: The CAST paper c

15 Nov 18, 2022
Learning to trade under the reinforcement learning framework

Trading Using Q-Learning In this project, I will present an adaptive learning model to trade a single stock under the reinforcement learning framework

Uirá Caiado 470 Nov 28, 2022
Pytorch implementations of Bayes By Backprop, MC Dropout, SGLD, the Local Reparametrization Trick, KF-Laplace, SG-HMC and more

Bayesian Neural Networks Pytorch implementations for the following approximate inference methods: Bayes by Backprop Bayes by Backprop + Local Reparame

1.4k Jan 07, 2023
PantheonRL is a package for training and testing multi-agent reinforcement learning environments.

PantheonRL is a package for training and testing multi-agent reinforcement learning environments. PantheonRL supports cross-play, fine-tuning, ad-hoc coordination, and more.

Stanford Intelligent and Interactive Autonomous Systems Group 57 Dec 28, 2022
GND-Nets (Graph Neural Diffusion Networks) in TensorFlow.

GNDC For submission to IEEE TKDE. Overview Here we provide the implementation of GND-Nets (Graph Neural Diffusion Networks) in TensorFlow. The reposit

Wei Ye 3 Aug 08, 2022
Learning Tracking Representations via Dual-Branch Fully Transformer Networks

Learning Tracking Representations via Dual-Branch Fully Transformer Networks DualTFR ⭐ We achieves the runner-ups for both VOT2021ST (short-term) and

phiphi 19 May 04, 2022
Small repo describing how to use Hugging Face's Wav2Vec2 with PyCTCDecode

🤗 Transformers Wav2Vec2 + PyCTCDecode Introduction This repo shows how 🤗 Transformers can be used in combination with kensho-technologies's PyCTCDec

Patrick von Platen 102 Oct 22, 2022
System-oriented IR evaluations are limited to rather abstract understandings of real user behavior

Validating Simulations of User Query Variants This repository contains the scripts of the experiments and evaluations, simulated queries, as well as t

IR Group at Technische Hochschule Köln 2 Nov 23, 2022
Taichi Course Homework Template

太极图形课S1-标题部分 这个作业未来或将是你的开源项目,标题的内容可以来自作业中的核心关键词,让读者一眼看出你所完成的工作/做出的好玩demo 如果暂时未想好,起名时可以参考“太极图形课S1-xxx作业” 如下是作业(项目)展开说明的方法,可以帮大家理清思路,并且也对读者非常友好,请小伙伴们多多参

TaichiCourse 30 Nov 19, 2022
Anomaly detection in multi-agent trajectories: Code for training, evaluation and the OpenAI highway simulation.

Anomaly Detection in Multi-Agent Trajectories for Automated Driving This is the official project page including the paper, code, simulation, baseline

12 Dec 02, 2022
A Model for Natural Language Attack on Text Classification and Inference

TextFooler A Model for Natural Language Attack on Text Classification and Inference This is the source code for the paper: Jin, Di, et al. "Is BERT Re

Di Jin 418 Dec 16, 2022
Calculates carbon footprint based on fuel mix and discharge profile at the utility selected. Can create graphs and tabular output for fuel mix based on input file of series of power drawn over a period of time.

carbon-footprint-calculator Conda distribution ~/anaconda3/bin/conda install anaconda-client conda-build ~/anaconda3/bin/conda config --set anaconda_u

Seattle university Renewable energy research 7 Sep 26, 2022