Code for Universal Semi-Supervised Semantic Segmentation models paper accepted in ICCV 2019

Tarun K

Last update: Nov 24, 2022

Related tags

Deep Learning machine-learning semi-supervised-learning deeplearning representation-learning semantic-segmentation domainadaptation iccv2019

Overview

USSS_ICCV19

Code for Universal Semi Supervised Semantic Segmentation accepted to ICCV 2019.
Full Paper available at https://arxiv.org/abs/1811.10323.

Requirements

Python >= 2.6
PyTorch >= 1.0.0
The ImageNet pretrained models are downloaded from the repository at https://github.com/fyu/drn.

Datasets

Cityscapes: https://www.cityscapes-dataset.com/
IDD: https://idd.insaan.iiit.ac.in/

How to run

python segment.py --basedir <basedir> --lr 0.001 --num-epochs 200 --batch-size 8 --savedir <savedir> --datasets <D1> [<D2> ..] --num-samples <N> --alpha 0 --beta 0 --resnet <resnet_v> --model drnet

Acknowledgements

Part of the code is heavily borrowed from the official code release of Dilated Residual Networks (https://github.com/fyu/drn) and IDD Dataset (https://github.com/AutoNUE/public-code).

Comments

About the unsupervised loss
First of all, thank you for this wonderful work and for providing the implementation

I am was interested in seeing how the unsupervised loss was implemented, in the paper, the loss is the entropy of sigmoid(v), the v is the similarity between the embedding of two labels set (either within or across the dataset).

I have two main questions if I may:

In the implementation I see that we only calculate the entropy between the element products of embeddings directly, is the sigmoid applied before the loss calculation because I see that it was omitted in this line. loss_unsup = {d:beta*torch.mean(-1* torch.sum(log_probs * tensor_prob , dim=1)).view(-1)}

I see that in the intialization, we initialize the weights of the embedding layer using torch.randn(NUM_LABELS , em_dim), does this help the training and it better than using the initiale weights of the conv layer in the beginning. And during training we don't update the weights of this conv1x1, can you please explain why is that.

Thank you.
opened by yassouali 4
Unsupervised Loss

Hi,

thank you for your contribution and for releasing your work.

I tried to rerun your model on Cityscape and IDD datasets. What I noticed particularly is that the Unsupervised Loss (US1, US2) is extremely low in comparison to typical loss. Did you put a weight vector on that Loss or is that a bug? Can you help me to understand this?

Epoch: 50 train loss : 0.7892 S1: 0.3645 S2: 0.3907 US1: 0.0006 US2:0.0004 Total: 0.7892 Best epoch : Val-IoU-CS= 0.2694 Val-IoU-IDD= 0.2173

Thank you Best wishes

opened by erdzl 1

Recall Loss for Semantic Segmentation (This repo implements the paper: Recall Loss for Semantic Segmentation)

Recall Loss for Semantic Segmentation (This repo implements the paper: Recall Loss for Semantic Segmentation) Download Synthia dataset The model uses

32 Sep 21, 2022

Official pytorch code for SSC-GAN: Semi-Supervised Single-Stage Controllable GANs for Conditional Fine-Grained Image Generation(ICCV 2021)

SSC-GAN_repo Pytorch implementation for 'Semi-Supervised Single-Stage Controllable GANs for Conditional Fine-Grained Image Generation'.PDF SSC-GAN:Sem

4 Aug 28, 2022

Pre-trained model, code, and materials from the paper "Impact of Adversarial Examples on Deep Learning Models for Biomedical Image Segmentation" (MICCAI 2019).

Adaptive Segmentation Mask Attack This repository contains the implementation of the Adaptive Segmentation Mask Attack (ASMA), a targeted adversarial

53 Jul 4, 2022

Learning Pixel-level Semantic Affinity with Image-level Supervision for Weakly Supervised Semantic Segmentation, CVPR 2018

Learning Pixel-level Semantic Affinity with Image-level Supervision This code is deprecated. Please see https://github.com/jiwoon-ahn/irn instead. Int

337 Dec 15, 2022

[CVPR 2021] MiVOS - Mask Propagation module. Reproduced STM (and better) with training code :star2:. Semi-supervised video object segmentation evaluation.

MiVOS (CVPR 2021) - Mask Propagation Ho Kei Cheng, Yu-Wing Tai, Chi-Keung Tang [arXiv] [Paper PDF] [Project Page] [Papers with Code] This repo impleme

106 Jan 3, 2023

Code for Universal Semi-Supervised Semantic Segmentation models paper accepted in ICCV 2019

Related tags

Overview

USSS_ICCV19

Requirements

Datasets

How to run

Acknowledgements

You might also like...

Recall Loss for Semantic Segmentation (This repo implements the paper: Recall Loss for Semantic Segmentation)

Official pytorch code for SSC-GAN: Semi-Supervised Single-Stage Controllable GANs for Conditional Fine-Grained Image Generation(ICCV 2021)

Pre-trained model, code, and materials from the paper "Impact of Adversarial Examples on Deep Learning Models for Biomedical Image Segmentation" (MICCAI 2019).

Learning Pixel-level Semantic Affinity with Image-level Supervision for Weakly Supervised Semantic Segmentation, CVPR 2018

Universal Adversarial Triggers for Attacking and Analyzing NLP (EMNLP 2019)

Code for the ICCV 2021 Workshop paper: A Unified Efficient Pyramid Transformer for Semantic Segmentation.

[ICCV-2021] An Empirical Study of the Collapsing Problem in Semi-Supervised 2D Human Pose Estimation

Adaptive Pyramid Context Network for Semantic Segmentation (APCNet CVPR'2019)

[CVPR 2021] MiVOS - Mask Propagation module. Reproduced STM (and better) with training code :star2:. Semi-supervised video object segmentation evaluation.

Comments

About the unsupervised loss

Unsupervised Loss

Owner

Tarun K

Shape-aware Semi-supervised 3D Semantic Segmentation for Medical Images

Semi-supervised Semantic Segmentation with Directional Context-aware Consistency (CVPR 2021)

Anti-Adversarially Manipulated Attributions for Weakly and Semi-Supervised Semantic Segmentation (CVPR 2021)

Semi-supervised Semantic Segmentation with Directional Context-aware Consistency (CVPR 2021)

[CVPR 2021] Semi-Supervised Semantic Segmentation with Cross Pseudo Supervision

ST++: Make Self-training Work Better for Semi-supervised Semantic Segmentation

[CVPR 2022] Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels

[cvpr22] Perturbed and Strict Mean Teachers for Semi-supervised Semantic Segmentation

A PyTorch implementation of "Semi-Supervised Graph Classification: A Hierarchical Graph Perspective" (WWW 2019)

Code release for Universal Domain Adaptation(CVPR 2019)