/awesome-weakly-supervised-semantic-segmentation

Awesome weakly-supervised image semantic segmentation；scribble supervision，bounding box supervision, point supervision, image-level class supervision

Awesome Weakly-supervised Semantic Segmentation

AwesomeGitHub stars GitHub forks visitors

Table of Contents

Contact gyguo95@gmail.com if any paper is missed!

1. Performance list

1.1. Results on PASCAL VOC 2012 dataset

  • For each method, I will provide the name of baseline in brackets if it has.
  • Sup.: I-image-level class label, B-bounding box label, S-scribble label, P-point label.
  • Bac. C: Method for generating pseudo label, or backbone of the classification network.
  • Arc. S: backbone and method of the segmentation network.
  • Pre.s : The dataset used to pre-train the segmentation network, "I" denotes ImageNet, "C" denotes COCO. Note that many works use COCO pre-trained DeepLab model but not mentioned in the paper.
  • For methods that use multiple backbones, I only reports the results of ResNet101.
  • "-" indicates no fully-supervised model is utilized, "?" indicates the corresponding item is not mentioned in the paper.

Image-level supervision with extra data

Method Pub. Bac. C Arc. S Sup. Extra data Pre.S val test
SEC ECCV16 VGG16 VGG16 DeepLabv1 I Saliency I 50.7 51.7
DSRG (SEC) CVPR18 VGG16 ResNet101 DeepLabv2 I Saliency I 61.4 63.2
AISI ECCV18 ResNet101 ResNet101 DeepLabv2 I Saliency ? 63.6 64.5
Ficklenet (DSRG) CVPR19 VGG16 ResNet101 DeepLabv2 I Saliency I 64.9 65.3
AISI ECCV18 ResNet101 ResNet101 DeepLabv2 I Saliency
24KImageNet		 ? 64.5 65.6
OAA ICCV19 VGG16 ResNet101 DeepLabv1 I Saliency I 65.2 66.4
Zhang et al. ECCV20 ResNet50 ResNet50 DeepLabv2 I Saliency ? 66.6 66.7
Fan et al. ECCV20 ResNet38 ResNet101 DeepLabv1 I Saliency ? 67.2 66.7
MCIS ECCV20 VGG16 ResNet101 DeepLabv1 I Saliency ? 66.2 66.9
Lee et al. ICCV19 VGG16 ResNet101 DeepLabv2 I Saliency Web I 66.5 67.4
LIID PAMI20 ResNet50 ResNet101 DeepLabv2 I Saliency ? 66.5 67.5
MCIS ECCV20 VGG16 ResNet101 DeepLabv1 I Saliency Web ? 67.7 67.5
ICD CVPR20 VGG16 ResNet101 DeepLabv1 I Saliency ? 67.8 68.0
LIID PAMI20 ResNet50 ResNet101 DeepLabv2 I Saliency
24KImageNet		 ? 67.8 68.3
Li et al. AAAI21 ResNet101 ResNet101 DeepLabv2 I Saliency ? 68.2 68.5
Yao et al. CVPR21 VGG16 ResNet101 DeepLabv2 I Saliency I 68.3 68.5
AuxSegNet ICCV21 ResNet38 - I Saliency ? 69.0 68.6
SPML (Ficklenet) ICLR21 VGG16 ResNet101 DeepLabv2 I Saliency I 69.5 71.6
Yao et al. CVPR21 VGG16 ResNet101 DeepLabv2 I Saliency I+C 70.4 70.2
WegFormer CVPR22 Deit-B ResNet101 DeepLabv ? I Saliency I 70.5 70.3
GETAM arxiv22 Deit-Distilled ResNet101 DeepLabv2 I Saliency I 70.6 70.4
WegFormer CVPR22 Deit-B ResNet101 DeepLabv ? I Saliency I+C 70.9 70.5
EDAM CVPR21 ResNet38 ResNet101 DeepLabv2 I Saliency ? 70.9 70.6
EPS CVPR21 ResNet38 ResNet101 DeepLabv2 I Saliency I 70.9 70.8
EPS CVPR21 ResNet38 ResNet101 DeepLabv1 I Saliency I 71.0 71.8
DRS AAAI21 VGG16 ResNet101 DeepLabv2 I Saliency I+C 71.2 71.4
ReCAM (EDAM) CVPR22 ResNet50 ResNet101 DeepLabv2 I Saliency I+C 71.8 72.2
L2G CVPR22 L2G ResNet101 DeepLabv1 I Saliency ? 72.0 73.0
L2G CVPR22 L2G ResNet101 DeepLabv2 I Saliency ? 72.1 71.7

Image-level supervision without extra data

Method Pub. Bac. C Arc. S Sup. Extra data Pre.S val test
AffinityNet CVPR18 ResNet38 ResNet38 I - ? 61.7 63.7
ICD CVPR20 VGG16 ResNet101 DeepLabv1 I - ? 64.1 64.3
IRN CVPR19 ResNet50 ResNet50 DeepLabv2 I - I 63.5 64.8
IAL IJCV20 ResNet? ResNet? I - I 64.3 65.4
SSDD (PSA) ICCV19 ResNet38 ResNet38 I - I 64.9 65.5
SEAM CVPR20 ResNet38 ResNet38 DeepLabv2 I - I 64.5 65.7
Chang et al. CVPR20 ResNet38 ResNet101 DeepLabv2 I - ? 66.1 65.9
RRM AAAI20 ResNet38 ResNet101 DeepLabv2 I - ? 66.3 66.5
BES ECCV20 ResNet50 ResNet101 DeepLabv2 I - ? 65.7 66.6
AFA CVPR22 MiT-B1 - I - ? 66.0 66.3
CONTA (+SEAM) NeurIPS20 ResNet38 ResNet101 DeepLabv2 I - ? 66.1 66.7
ESC-Net ICCV21 ResNet38 ResNet38 DeepLabv2 I - I 66.6 67.6
Ru et al. IJCAI21 ResNet101 ResNet101 DeepLabv2 I - ? 67.2 67.3
WSGCN (IRN) ICME21 ResNet50 ResNet101 DeepLabv2 I - I 66.7 68.8
CPN ICCV21 ResNet38 ResNet38 DeepLabv1 I - ? 67.8 68.5
RPNet TMM21 ResNet101 ResNet50 DeepLabv2 I - I 68.0 68.2
AdvCAM CVPR21 ResNet50 ResNet101 DeepLabv2 I - I 68.1 68.0
ReCAM CVPR22 ResNet50 ResNet101 DeepLabv2 I - I 68.5 68.4
PMM ICCV21 ResNet38 ResNet38 PSPnet I - ? 68.5 69.0
WSGCN (IRN) ICME21 ResNet50 ResNet101 DeepLabv2 I - I+C 68.7 69.3
ASDT arxiv22 ResNet38 ResNet101 DeepLabv2 I - I 69.7 70.1
PMM ICCV21 Res2Net101 Res2Net101 PSPnet I - ? 70.0 70.5
ASDT arxiv22 ResNet38 Res2Net101 PSPnet I - I 71.1 71.0
MCTformer CVPR22 DeiT-S ResNet38 DeeplabV1 I - ? 71.9 71.6

Box-level supervision

Method Pub. Bac. C Arc. S Sup. Extra data Pre.S val test
BBAM CVPR21 ? ResNet101 DeepLabv2 B MCG I 73.7 73.7
WSSL ICCV15 - VGG16 DeepLabv1 B - I 60.6 62.2
Song et al. CVPR19 - ResNet101 DeepLabv1 B - I 70.2 -
SPML (Song et al.) ICLR21 - ResNet101 DeepLabv2 B - I 73.5 74.7
Oh et al. CVPR21 ResNet101 ResNet101 DeepLabv2 B - I+C 74.6 76.1

Scribble-level supervision

Method Pub. Bac. C Arc. S Sup. Extra data Pre.S val test
Scribblesup CVPR16 - VGG16 DeepLabv1 S - ? 63.1 -
NormalCut CVPR18 - ResNet101 DeepLabv1 S Saliency ? 74.5 -
KernelCut ECCV18 - ResNet101 DeepLabv1 S - ? 75.0 -
BPG IJCAI19 - ResNet101 DeepLabv2 S - ? 76.0 -
SPML (KernelCut) ICLR21 - ResNet101 DeepLabv2 S - I 76.1 -
A2GNN TPAMI21 - ? S - ? 76.2 76.1
DFR arxiv21 - UperNet+Swin Transformer S 22KImageNet - 82.8 82.9

Point-level supervision

Method Pub. Bac. C Arc. S Sup. Extra data Pre.S val test
WhatsPoint ECCV16 - VGG16 FCN P Objectness I 46.1 -
PCAM arxiv20 ResNet50 DeepLabv3+ P - ? 70.5 -

1.2. Results on MS-COCO dataset

Image-level supervision with extra data

Method Pub. Bac. C Arc. S Sup. Extra data val test
AuxSegNet ICCV21 ResNet38 - I Saliency 33.9 -
EPS CVPR21 ResNet38 ResNet101 DeepLabv2 I Saliency 35.7 -
L2G CVPR22 L2G VGG16 DeepLabv2 I Saliency 42.7 -
L2G CVPR22 L2G ResNet101 DeepLabv2 I Saliency 44.2 -

Image-level supervision without extra data

Method Pub. Bac. C Arc. S Sup. Extra data val test
MCTformer CVPR22 DeiT-S ResNet38 DeeplabV1 I - 42.0 -
ReCAM (AdvCAM + IRN) CVPR22 ResNet50 ResNet101 DeepLabv2 I - 45.0 -

2. Paper List

2.1. supervised by image tags (I)

2022

  • MCTformer: Multi-class Token Transformer for Weakly Supervised Semantic Segmentation CVPR2022
  • AFA: Learning Affinity from Attention End-to-End Weakly-Supervised Semantic Segmentation with Transformers CVPR2022
  • WegFormer: WegFormer Transformers for Weakly Supervised Semantic Segmentation CVPR2022
  • L2G: L2G: A Simple Local-to-Global Knowledge Transfer Framework for Weakly Supervised Semantic Segmentation CVPR2022
  • ReCAM: Class Re-Activation Maps for Weakly-Supervised Semantic Segmentation. CVPR2022
  • GETAM: GETAM: Gradient-weighted Element-wise Transformer Attention Map for Weakly-supervised Semantic segmentation arxiv2022
  • ASDT: Weakly Supervised Semantic Segmentation via Alternative Self-Dual Teaching arxiv2022

2021

  • SPML: "Universal Weakly Supervised Segmentation by Pixel-to-Segment Contrastive Learning" ICLR2021
  • Li et al.: "Group-Wise Semantic Mining for Weakly Supervised Semantic Segmentation" AAAI2021
  • DRS: "Discriminative Region Suppression for Weakly-Supervised Semantic Segmentation" AAAI2021
  • AdvCAM: " Anti-Adversarially Manipulated Attributions for Weakly and Semi-Supervised Semantic Segmentation" CVPR2021
  • **Yao et al. **: "Non-Salient Region Object Mining for Weakly Supervised Semantic Segmentation" CVPR2021
  • EDAM: "Embedded Discriminative Attention Mechanism for Weakly Supervised Semantic Segmentation" CVPR2021
  • EPS: Railroad is not a Train Saliency as Pseudo-pixel Supervision for Weakly Supervised Semantic Segmentation CVPR2021
  • WSGCN: "Weakly-Supervised Image Semantic Segmentation Using Graph Convolutional Networks" ICME2021
  • PuzzleCAM: "Puzzle-CAM Improved localization via matching partial and full features" 2021arXiv
  • CDA: "Context Decoupling Augmentation for Weakly Supervised Semantic Segmentation" ICCV2021
  • ECS-Net: ECS-Net: Improving Weakly Supervised Semantic Segmentation by Using Connections Between Class Activation Maps.* ICCV2021*
  • Ru et al.: "Learning Visual Words for Weakly-Supervised Semantic Segmentation" IJCAI2021
  • AuxSegNet: "Leveraging Auxiliary Tasks with Affinity Learning for Weakly Supervised Semantic Segmentation" ICCV2021
  • CPN: "Complementary Patch for Weakly Supervised Semantic Segmentation" ICCV2021
  • PMM: "Pseudo-mask Matters in Weakly-supervised Semantic Segmentation" ICCV2021
  • RPNet: "Cross-Image Region Mining with Region Prototypical Network for Weakly Supervised Segmentation" TMM2021
  • Weakly-supervised semantic segmentation with superpixel guided local and global consistency PR2021

2020

  • RRM: "Reliability Does Matter An End-to-End Weakly Supervised Semantic Segmentation Approach" AAAI2020
  • IAL: "Weakly-Supervised Semantic Segmentation by Iterative Affinity Learning" IJCV2020
  • SEAM: "Self-supervised Equivariant Attention Mechanism for Weakly Supervised Semantic Segmentation" CVPR2020
  • Chang et al.: "Weakly-Supervised Semantic Segmentation via Sub-category Exploration" CVPR2020
  • ICD: "Learning Integral Objects with Intra-Class Discriminator for Weakly-Supervised Semantic Segmentation" CVPR2020
  • Fan et al.: "Employing multi-estimations for weakly-supervised semantic segmentation" ECCV2020
  • MCIS: "Mining Cross-Image Semantics for Weakly Supervised Semantic Segmentation" 2020
  • BES: "Weakly Supervised Semantic Segmentation with Boundary Exploration" ECCV2020
  • CONTA: "Causal intervention for weakly-supervised semantic segmentation" NeurIPS2020
  • Method: "Find it if You Can: End-to-End Adversarial Erasing for Weakly-Supervised Semantic Segmentation" 2020arXiv
  • Zhang et al.: "Splitting vs. Merging: Mining Object Regions with Discrepancy and Intersection Loss for Weakly Supervised Semantic Segmentation" ECCV2020
  • LIID "Leveraging Instance-, Image- and Dataset-Level Information for Weakly Supervised Instance Segmentation" TPAMI2020

2019

  • IRN: "Weakly Supervised Learning of Instance Segmentation with Inter-pixel Relations" CVPR2019
  • Ficklenet: " Ficklenet: Weakly and semi-supervised semantic image segmentation using stochastic inference" CVPR2019
  • Lee et al.: "Frame-to-Frame Aggregation of Active Regions in Web Videos for Weakly Supervised Semantic Segmentation" ICCV2019
  • OAA: "Integral Object Mining via Online Attention Accumulation" ICCV2019
  • SSDD: "Self-supervised difference detection for weakly-supervised semantic segmentation" ICCV2019

2018

  • DSRG: "Weakly-supervised semantic segmentation network with deep seeded region growing" CVPR2018
  • AffinityNet: "Learning Pixel-level Semantic Affinity with Image-level Supervision for Weakly Supervised Semantic Segmentation" CVPR2018
  • GAIN: " Tell me where to look: Guided attention inference network" CVPR2018
  • AISI: "Associating inter-image salient instances for weakly supervised semantic segmentation" ECCV2018
  • SeeNet: "Self-Erasing Network for Integral Object Attention" NeurIPS2018
  • Method: "" 2018

2017

  • CrawlSeg: "Weakly Supervised Semantic Segmentation using Web-Crawled Videos" CVPR2017
  • WebS-i2: "Webly supervised semantic segmentation" CVPR2017
  • Oh et al.: "Exploiting saliency for object segmentation from image level labels" CVPR2017
  • TPL: "Two-phase learning for weakly supervised object localization" ICCV2017

2016

  • SEC: "Seed, expand and constrain: Three principles for weakly-supervised image segmentation" ECCV2016
  • AF-SS: "Augmented Feedback in Semantic Segmentation under Image Level Supervision" 2016
  • DCSM: Distinct class-specific saliency maps for weakly supervised semantic segmentation ECCV2016

2.2. Supervised by bounding box (B)

  • WSSL: "Weakly-and semi-supervised learning of a deep convolutional network for semantic image segmentation" ICCV2015
  • Boxsup: "Boxsup: Exploiting bounding boxes to supervise convolutional networks for semantic segmentation" ICCV2015
  • Song et al.: "Box-driven class-wise region masking and filling rate guided loss for weakly supervised semantic segmentation" CVPR2019
  • BBAM: "BBAM: Bounding Box Attribution Map for Weakly Supervised Semantic and Instance Segmentation" CVPR2021
  • Oh et al.: "Ba ckground-Aware Pooling and Noise-Aware Loss for Weakly-Supervised Semantic Segmentation" CVPR2021
  • SPML: "Universal Weakly Supervised Segmentation by Pixel-to-Segment Contrastive Learning" ICLR2021

2.3. Supervised by scribble (S)

  • Scribblesup: "Scribblesup: Scribble-supervised convolutional networks for semantic segmentation" CVPR2016
  • NormalCut : "Normalized cut loss for weakly-supervised cnn segmentation" CVPR2018
  • KernelCut : "On regularized losses for weakly-supervised cnn segmentation" ECCV2018
  • BPG: "Boundary Perception Guidance: A Scribble-Supervised Semantic Segmentation Approach" IJCAI2019
  • SPML: "Universal Weakly Supervised Segmentation by Pixel-to-Segment Contrastive Learning" ICLR2021
  • DFR: "Dynamic Feature Regularized Loss for Weakly Supervised Semantic Segmentation" arxiv2021
  • A2GNN: "Affinity attention graph neural network for weakly supervised semantic segmentation" TPAMI2021

2.4. Supervised by point (P)

  • WhatsPoint: "What’s the Point: Semantic Segmentation with Point Supervision" ECCV2016
  • PCAM: "PCAMs: Weakly Supervised Semantic Segmentation Using Point Supervision" arxiv2020

3. Dataset

PASCAL VOC 2012

@article{everingham2010pascal,
  title={The pascal visual object classes (voc) challenge},
  author={Everingham, Mark and Van Gool, Luc and Williams, Christopher KI and Winn, John and Zisserman, Andrew},
  journal={International journal of computer vision},
  volume={88},
  number={2},
  pages={303--338},
  year={2010},
  publisher={Springer}
}

MS COCO 2014

@inproceedings{lin2014microsoft,
  title={Microsoft coco: Common objects in context},
  author={Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll{\'a}r, Piotr and Zitnick, C Lawrence},
  booktitle={European conference on computer vision},
  pages={740--755},
  year={2014},
  organization={Springer}
}