Skip to content

Latest commit

 

History

History
409 lines (206 loc) · 68.7 KB

Semi-Supervised_Learning.md

File metadata and controls

409 lines (206 loc) · 68.7 KB
Raw

Contents

Materials

Pioneers

[李宇峰 Yu-Feng Li]

Papers (Semi-Supervised_Learning)

▶for Considering Domain Adaptation

  • (TPAMI2018) Semi-Supervised Domain Adaptation by Covariance Matching [paper link][Xi’an Jiaotong University]

  • 👍SSDA_MME(ICCV2019) Semi-supervised Domain Adaptation via Minimax Entropy [paper link][arxiv link][project link][code|official][Boston University and University of California, Berkeley]

  • ECACL(ICCV2021) ECACL: A Holistic Framework for Semi-Supervised Domain Adaptation [paper link][arxiv link][code|official][NEC Laboratories, America and Northeastern University][the code is based on SSDA_MME]

▶for Data Augmentation

▶for Image Classification

  • Π-model (ICLR2017) Temporal Ensembling for Semi-Supervised Learning [openreview link][arxiv link][It proposes a temporal ensemble strategy for the pseudo-label to reduce the noise in the target]

  • 👍👍Mean Teachers (NIPS2017) Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results [paper link][arxiv link][ the teacher is the moving average of the student which can be timely updated in every iteration, But their performance is limited because the two models tend to converge to the same point and stop further exploration]

  • VAT (TPAMI2018) Virtual Adversarial Training: A Regularization Method for Supervised and Semi-Supervised Learning [paper link][arxiv link][code|official vat_chainer][code|official vat_tf]

  • DCT (Deep Co-Training)(ECCV2018) Deep Co-Training for Semi-Supervised Image Recognition [paper link][arxiv link][learn two different models by minimizing their prediction discrepancy, learn from different initializations to avoid the case where the two models converge to the same point]

  • Dual-Student (ICCV2019) Dual Student: Breaking the Limits of the Teacher in Semi-Supervised Learning [paper link][learn two different models by minimizing their prediction discrepancy, add view difference constraints to avoid the case where the two models converge to the same point]

  • 👍MixMatch (NIPS2019) MixMatch: A Holistic Approach to Semi-Supervised Learning [paper link][arxiv link][code|official][Google, The first author is David Berthelot, Combining Existing Useful SSL Techniques]

  • 👍ReMixMatch (ICLR2020) ReMixMatch: Semi-Supervised Learning with Distribution Matching and Augmentation Anchoring [openreview link][arxiv link][code|official][Google, The first author is David Berthelot, Applying Multiple Strong Augmentations for the Same Input Batch]

  • 👍👍FixMatch (NIPS2020) FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence [paper link][arxiv link][code|official][Google, The first author is David Berthelot, Weak-Strong Augmentation Pairs, pseudo-labeling based (also called self-training)]

  • FeatMatch(ECCV2020) FeatMatch: Feature-Based Augmentation for Semi-supervised Learning [paper link][arxiv link][project link][code|official]

  • 👍UDA (NIPS2020) Unsupervised Data Augmentation for Consistency Training [paper link][arxiv link][code|official]

  • 👍FlexMatch (NIPS2021) FlexMatch: Boosting Semi-Supervised Learning with Curriculum Pseudo Labeling [paper link][arxiv link][code|official]

  • Dash (ICML2021) Dash: Semi-Supervised Learning with Dynamic Thresholding [paper link][arxiv link][It proposes dynamic and adaptive pseudo label filtering, better suited for the training process (similar to the FixMatch)]

  • SimPLE (CVPR2021) SimPLE: Similar Pseudo Label Exploitation for Semi-Supervised Classification [paper link][arxiv link][code|official][It proposes the paired loss minimizing the statistical distance between confident and similar pseudo labels]

  • SemCo (CVPR2021) All Labels Are Not Created Equal: Enhancing Semi-Supervision via Label Grouping and Co-Training [paper link][arxiv link][code|official][It considers label semantics to prevent the degradation of pseudo label quality for visually similar classes in a co-training manner]

  • EMAN (CVPR2021) Exponential Moving Average Normalization for Self-Supervised and Semi-Supervised Learning [paper link][arxiv link][code|official][may not that generic]

  • CoMatch (ICCV2021) CoMatch: Semi-Supervised Learning With Contrastive Graph Regularization [paper link][arxiv link][code|official][Salesforce Research, based on `FixMatch]

  • CRMatch (arxiv2021.12)(IJCV2023) Revisiting Consistency Regularization for Semi-Supervised Learning [paper link][arxiv link]

  • SAW_SSL(ICML2022) Smoothed Adaptive Weighting for Imbalanced Semi-Supervised Learning: Improve Reliability Against Unknown Distribution Data [paper link][code|official]

  • ADSH(ICML2022) Class-Imbalanced Semi-Supervised Learning with Adaptive Thresholding [paper link][code|official][Nanjing University]

  • Classification-SemiCLS (CVPR2022) Class-Aware Contrastive Semi-Supervised Learning [paper link][arxiv link][code|official][based on FixMatch, THU + Tencent Youtu Lab]

  • SimMatch (CVPR2022) SimMatch: Semi-Supervised Learning With Similarity Matching [paper link][arxiv link][code|official][The University of Sydney]

  • USB(NIPS2022) USB: A Unified Semi-supervised Learning Benchmark for Classification [paper link][arxiv link][code|official][microsoft]

  • Bi-Adaptation(ICML2023) Bidirectional Adaptation for Robust Semi-Supervised Learning with Inconsistent Data Distributions [openreview link][pdf link][Nanjing University, SSL (Semi-Supervised Learning) + DA (Domain Adaptation)]

  • FreeMatch (ICLR2023) FreeMatch: Self-adaptive Thresholding for Semi-supervised Learning [openreview link][arxiv link][code|official][microsoft]

  • SoftMatch (ICLR2023) SoftMatch: Addressing the Quantity-Quality Trade-off in Semi-supervised Learning [openreview link][arxiv link][code|official][microsoft]

  • FullMatch (CVPR2023) Boosting Semi-Supervised Learning by Exploiting All Unlabeled Data [paper link][arxiv link][code|official][based on FixMatch, megvii-research]

  • CHMatch (CVPR2023) CHMATCH: Contrastive Hierarchical Matching and Robust Adaptive Threshold Boosted Semi-Supervised Learning [paper link][code|official][based on FixMatch and FlexMatch, Harbin Institute of Technology (Shenzhen)]

  • 👍Suave-Daino(CVPR2023) Semi-Supervised Learning Made Simple With Self-Supervised Clustering [paper link][arxiv link][code|official][based on Self-Supervised methods such as SwAV or DINO]

  • ProtoCon(CVPR2023) PROTOCON: Pseudo-label Refinement via Online Clustering and Prototypical Consistency for Efficient Semi-supervised Learning [paper link][arxiv link][code is unavailable][Monash University, Australia]

  • SAA (ICCV2023) Enhancing Sample Utilization through Sample Adaptive Augmentation in Semi-Supervised Learning [paper link][arxiv link][code|official][based on FixMatch and FlexMatch]

  • ShrinkMatch (ICCV2023) Shrinking Class Space for Enhanced Certainty in Semi-Supervised Learning [paper link][arxiv link][code|official][based on FixMatch]

  • SimMatchV2 (ICCV2023) SimMatchV2: Semi-Supervised Learning with Graph Consistency [paper link][arxiv link][code|official][The University of Sydney, based on FixMatch and SimMatch]

  • **** [][][]

▶for Object Detection

▶for Semantic Segmentation

▶for Pose Estimation

▶for 3D Object Detection

It estimates the category and 3D bounding box for each object in the image. The 3D bounding box can be further divided into 3D center location (x, y, z), dimension (h, w, l) and orientation (yaw angle) θ. The roll and pitch angles of objects are set to 0.

  • 👍SESS (CVPR2020 oral) SESS: Self-Ensembling Semi-Supervised 3D Object Detection [paper link][arxiv link][code|official][National University of Singapore, comparing to SESS, on datasets ScanNet and SUNRGB-D, using Mean-Teacher]

  • 👍👍3DIoUMatch (CVPR2021) 3DIoUMatch: Leveraging IoU Prediction for Semi-Supervised 3D Object Detection [paper link][arxiv link][code|official][Stanford University + Tsinghua University + NVIDIA, on datasets ScanNet, SUNRGB-D and KITTI, using Mean-Teacher and FixMatch]

  • MVC-MonoDet (ECCV2022) Semi-supervised Monocular 3D Object Detection by Multi-view Consistency [paper link][code}official][3D detection on the KITTI and nuScenes datasets]

  • Proficient-Teachers (ECCV2022) Semi-supervised 3D Object Detection with Proficient Teachers [paper link][arxiv link][code|official (not really released!!!👎)][Beijing Institute of Technology, comparing to SESS and 3DIoUMatch yet no code, on datasets ONCE and Waymo Open, revised based on Mean-Teacher]

  • UpCycling (ICCV2023) UpCycling: Semi-supervised 3D Object Detection without Sharing Raw-level Unlabeled Scenes [paper link][arxiv link][Seoul National University, comparing to 3DIoUMatch yet no code, on datasets ScanNet, SUNRGB-D and KITTI]

  • ViT-WSS3D (ICCV2023) A Simple Vision Transformer for Weakly Semi-supervised 3D Object Detection [paper link][HUST, comparing to 3DIoUMatch yet no code, on datasets SUNRGB-D and KITTI]

  • Side-Aware (ICCV2023) Not Every Side Is Equal: Localization Uncertainty Estimation for Semi-Supervised 3D Object Detection [paper link][USTC, comparing to SESS and 3DIoUMatch yet no code, using Mean-Teacher]

  • NoiseDet (ICCV2023) Learning from Noisy Data for Semi-Supervised 3D Object Detection [paper link][code|official (not really released!!!👎)][USTC, comparing to SESS and 3DIoUMatch yet no code, on datasets ONCE and Waymo Open]

  • 👍👍DQS3D (ICCV2023) DQS3D: Densely-matched Quantization-aware Semi-supervised 3D Detection [paper link][arxiv link][code|official][Institute for AI Industry Research (AIR), Tsinghua University, comparing to SESS and 3DIoUMatch, on datasets ScanNet and SUNRGB-D, using Mean-Teacher]

▶for 6D Object Pose Estimation

  • multipath(CVPR2020) Multi-Path Learning for Object Pose Estimation Across Domains [paper link][code|official][Domain Adaptation, 6D Object Detection, 3D Object Pose Estimation]

  • Self6D(ECCV2020) Self6D: Self-Supervised Monocular 6D Object Pose Estimation [paper link][arxiv link][code|official (Self6D-Diff-Renderer)][THU]

  • Self6D++(TPAMI2021) Occlusion-Aware Self-Supervised Monocular 6D Object Pose Estimation [paper link][arxiv link][code|official][THU]

  • DSC-PoseNet(CVPR2021) DSC-PoseNet: Learning 6DoF Object Pose Estimation via Dual-scale Consistency [paper link][arxiv link][code is not available][Baidu Research, a self-supervised manner, needing 2D bounding boxes]

  • zero-shot-pose(ECCV2022) Zero-Shot Category-Level Object Pose Estimation [paper link][arxiv link][code|official][Zero-Shot Learning, University of Oxford, on the dataset CO3D, the authors re-annotated 10 sequences from each of 20 categories with ground-truth poses., all baselines are reproduced by the authors.]

  • Self-DPDN(ECCV2022) Category-Level 6D Object Pose and Size Estimation Using Self-supervised Deep Prior Deformation Networks [paper link][arxiv link][code|official][Self-Supervised, Domain Adaptation, South China University of Technology, another work VI-Net (ICCV2023) with title VI-Net: Boosting Category-level 6D Object Pose Estimation via Learning Decoupled Rotations on the Spherical Representations]

  • 👍Wild6D + RePoNet (NIPS2022) Category-Level 6D Object Pose Estimation in the Wild: A Semi-Supervised Learning Approach and A New Dataset [paper link][arxiv link][project link][code|official][University of California San Diego, a new dataset Wild6D, Xiaolong Wang, Yang Fu 付旸, based on the NOCS]

  • UDA-COPE(CVPR2022) UDA-COPE: Unsupervised Domain Adaptation for Category-level Object Pose Estimation [paper link][arxiv link][project link][no code][Domain Adaptation, Taeyeop Lee, based on the NOCS]

  • SSC-6D(AAAI2022) Self-Supervised Category-Level 6D Object Pose Estimation with Deep Implicit Shape Representation [paper link][code|official][Dalian University of Technology]

  • MAST(IJCAI2023) Manifold-Aware Self-Training for Unsupervised Domain Adaptation on Regressing 6D Object Pose [paper link][arxiv link][Domain Adaptation, Self-Training]

  • 👍self-pose(ICLR2023)(arxiv 2022.10) Self-Supervised Geometric Correspondence for Category-Level 6D Object Pose Estimation in the Wild [openreview link][arxiv link][project link][code|official][training and testing on Wild6D, Kaifeng Zhang, second author is Yang Fu 付旸]

  • TTA-COPE (CVPR2023) TTA-COPE: Test-Time Adaptation for Category-Level Object Pose Estimation [paper link][arxiv link][project link][Code is not available][Test-Time Adaptation, Taeyeop Lee, The proposed pose ensemble and the self-training loss improve category-level object pose performance during test time under both semi-supervised and unsupervised settings., based on the NOCS]

  • 👍PseudoFlow(ICCV2023) Pseudo Flow Consistency for Self-Supervised 6D Object Pose Estimation [paper link][arxiv link][code|official][Yang Hai(海洋), Yinlin Hu (胡银林)]

  • SA6D(CoRL2023) SA6D: Self-Adaptive Few-Shot 6D Pose Estimator for Novel and Occluded Objects [openreview link][project link][arxiv link][Bosch Center for AI, robotic manipulation, few-shot pose estimation (FSPE), inputs: a small number of cluttered reference images]

  • Cas6D (arxiv2023.06) Learning to Estimate 6DoF Pose from Limited Data: A Few-Shot, Generalizable Approach using RGB Images [arxiv link][Few-Shot Learning, ByteDance, compared to OnePose++ and Gen6D, trained on datasets LINEMOD and GenMOP, inputs: sparse support views]

▶for Rotation Regression (3D Object Pose)

We here also collect supervised learning based Rotation Regression (a.k.a. 3D Object Pose, or Camera Viewpoint Estimation) methods

Supervised Learning

  • ViewpointsAndKeypoints(CVPR2015) Viewpoints and Keypoints [paper link][arxiv link][code|official][University of California, Berkeley, the first author Shubham Tulsiani]

  • RenderForCNN(ICCV2015 Oral) Render for cnn: Viewpoint estimation in images using cnns trained with rendered 3d model views [paper link][arxiv link][project link][code|official][Stanford University, tested on dataset PASCAL3D+, proposed the synthesized dataset RenderForCNN]

  • deep_direct_stat(ECCV2018) Deep Directional Statistics: Pose Estimation with Uncertainty Quantification [paper link][arxiv link][code|official][MPII, von Mises, first author Sergey Prokudin, Probabilistic representations have been introduced for modeling orientation with uncertainty]

  • StarMap(ECCV2018) StarMap for Category-Agnostic Keypoint and Viewpoint Estimation [paper link][arxiv link][code|official][The University of Texas at Austin, tested on datasets ObjectNet3D and PASCAL3D+]

  • multi-modal-regression(BMVC2018) A Mixed Classification-Regression Framework for 3D Pose Estimation from 2D Images [pdf link][arxiv link][code|official][JHUVisionLab][used extra rendered images in RenderForCNN (ICCV2015) for training models on PASCAL3D+, then methods Spherical_Regression(CVPR2019), matrixFisher(NIPS2020), Implicit-PDF(ICML2021), RotationNormFlow(CVPR2023), Image2Sphere(ICLR2023), RotationLaplace(ICLR2023) and RestrictedRepresentations(NIPS2023) followed this setting]

  • Kpt+PnP(CVPR2018) 3D Pose Estimation and 3D Model Retrieval for Objects in the Wild [paper link][arxiv link][Graz University of Technology, Austria, tested on dataset PASCAL3D+][no code]

  • PoseFromShape(BMVC2019) Pose from Shape: Deep Pose Estimation for Arbitrary 3D Objects [arxiv link][project link][code|official][tested on datasets ObjectNet3D, Pascal3D+ and Pix3D][the first author Yang Xiao (肖洋)]

  • spherical_embeddings(ICML2019) Cross-Domain 3D Equivariant Image Embeddings [paper link][arxiv link][code|official][University of Pennsylvania + Google Research, on datasets ShapeNet, ModelNet and PASCAL3D+][It points out that Spherical CNNs are equivariant to 3D rotations][also see Implicit-PDF(ICML2021), Image2Sphere(ICLR2023) and RestrictedRepresentations(NIPS2023) which are partly inspired by this work.]

  • 👍 (CVPR2019) On the Continuity of Rotation Representations in Neural Networks [paper link][arxiv link][University of Southern California, the first author Yi Zhou, Gram-Schmidt orthogonalization procedure][For Rotation Regression, this paper validated that parameterization in four or fewer dimensions will be discontinuous (this applies to all classic representations such as Euler angles, axis-angle, and unit quaternions)]

  • 👍Spherical_Regression(CVPR2019) Spherical Regression: Learning Viewpoints, Surface Normals and 3D Rotations on N-Spheres [paper link][arxiv link][code|official][University of Amsterdam, the first author Shuai Liao, proposed the dataset ModelNet10-SO3]

  • deep_bingham(ICLR2020) Deep Orientation Uncertainty Learning based on a Bingham Loss [openreview link][code|official][MIT, Bingham distributions, first author Igor Gilitschenski, Probabilistic representations have been introduced for modeling orientation with uncertainty]

  • bingham-rotation-learning(QCQP)(RSS2020)(Best Student Paper Award Winner) A Smooth Representation of Belief over SO(3) for Deep Rotation Learning with Uncertainty [paper link][arxiv link][project link][code|official][University of Toronto + MIT, first author Valentin Peretroukhin][The unit quaternion that best aligns two point sets can be computed via the eigendecomposition of a symmetric data matrix, and the proposed network model regresses directly the elements of this 4x4 symmetric matrix.]

  • DeepBinghamNetworks(arxiv2020.12)(IJCV2022) Deep Bingham Networks: Dealing with Uncertainty and Ambiguity in Pose Estimation [paper link][arxiv link][project link][code|official][Stanford University + Technical University of Munich][The extended version of paper Multimodal Inference for 6D Camera Relocalization and Object Pose Estimation (ECCV2020)]

  • 👍 (NIPS2020) An Analysis of SVD for Deep Rotation Estimation [paper link][arxiv link][Simon Fraser University, the first author Jake Levinson, FisherMatch is partially based on the matrix Fisher distribution theory introduced in this paper.][pointed out two previous methods modeling the uncertainty of 3D rotation estimation: deep_direct_stat(ECCV2018) and deep_bingham(ICLR2020).]

  • 👍 (NIPS2020) Probabilistic Orientation Estimation with Matrix Fisher Distributions [paper link][arxiv link][code|official][KTH, the first author David Mohlin, FisherMatch is based on the matrix Fisher distribution theory introduced in this paper.][matrix Fisher distribution --> (1977) The von Mises–Fisher Matrix Distribution in Orientation Statistics][The visualiztion is adopted from --> (TAC2018) Bayesian Attitude Estimation with the Matrix Fisher Distribution on SO(3) (arxiv link)][pointed out two previous methods modeling the uncertainty of 3D rotation estimation: deep_direct_stat(ECCV2018) and deep_bingham(ICLR2020).][applying the matrix Fisher parameters for Human Mesh Recovery in work (CVPR2023) Learning Analytical Posterior Probability for Human Mesh Recovery]

  • PoseContrast(3DV2021 Oral) PoseContrast: Class-Agnostic Object Viewpoint Estimation in the Wild with Pose-Aware Contrastive Learning [paper link][arxiv link][project link][code|official][tested on datasets ObjectNet3D, Pascal3D+ and Pix3D][the first author Yang Xiao (肖洋)]

  • NeMo(ICLR2021) NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation [openreview link][arxiv link][code|official][Johns Hopkins University, tested on datasets ObjectNet3D, OccludedPASCAL3D+ and PASCAL3D+][the first author Angtian Wang]

  • 👍Implicit-PDF(ICML2021) Implicit-PDF: Non-Parametric Representation of Probability Distributions on the Rotation Manifold [paper link][arxiv link][project link][code|official][Google Research, the first author Kieran A Murphy , awesome visualization of SO(3)][The technical design choices of implicit pose model in this paper are inspired by the very successful implicit shape (OccupancyNetworks) and scene (NeRF) representations, which can represent detailed geometry with a multilayer perceptron that takes low-dimensional position and/or directions as inputs.][IPDF is quite good at outputting multiple 3D pose candidates for symmetry objects.]

  • RPMG(CVPR2022) Projective Manifold Gradient Layer for Deep Rotation Regression [paper link][arxiv link][project link][code|official][PKU, the first author Jiayi Chen | 陈嘉毅]

  • AcciTurn(ICCVW2023) Accidental Turntables: Learning 3D Pose by Watching Objects Turn [paper link][arxiv link][project link][dataset link][University of Massachusetts - Amherst + Adobe Research]

  • RotationNormFlow(CVPR2023) Delving into Discrete Normalizing Flows on SO (3) Manifold for Probabilistic Rotation Modeling [paper link][arxiv link][project link][code|official][PKU, the first author Yulin Liu 刘雨霖, it is also proposed by the co-author of FisherMatch]

  • Image2Sphere(ICLR2023)(notable-top-5%) Image to Sphere: Learning Equivariant Features for Efficient Pose Prediction [openreview link][arxiv link][project link][code|official][Northeastern University, the first author David M. Klee]

  • RotationLaplace(ICLR2023)(notable-top-25%) A Laplace-inspired Distribution on SO(3) for Probabilistic Rotation Estimation [openreview link][arxiv link][project link][code|official][PKU, the first author Yingda Yin 尹英达]

  • VoGE(ICLR2023)(Poster) VoGE: A Differentiable Volume Renderer using Gaussian Ellipsoids for Analysis-by-Synthesis [openreview link][arxiv link][code|official][following the Analysis-by-Synthesi pattern as NeMo][Johns Hopkins University, tested on datasets ObjectNet3D, OccludedPASCAL3D+ and PASCAL3D+][the first author Angtian Wang]

  • RestrictedRepresentations(NIPS2023)(arxiv2023.06) Equivariant Single View Pose Prediction Via Induced and Restricted Representations [openreview link][arxiv link][Northeastern University, the first author Owen Howell, on datasets PASCAL3D+ and SYMSOL, Machine Learning, Group Theory]

  • RnC(NIPS2023, Spotlight)(arxiv2022.10) Rank-N-Contrast: Learning Continuous Representations for Regression [paper link][openreview link][arxiv link][code|official][MIT CSAIL + GIST]

  • (arxiv2024.04) Learning a Category-level Object Pose Estimator without Pose Annotations [arxiv link][Xi’an Jiaotong Univeristy + Johns Hopkins University + Tsinghua University + University of Freiburg + MPII][based on Zero-1-to-3; tested on datasets PASCAL3D+ and KITTI]

Few-Shot Learning

  • FSDetView(ECCV2020) Few-Shot Object Detection and Viewpoint Estimation for Objects in the Wild [paper link][arxiv link][project link][tested on datasets ObjectNet3D and Pascal3D+]

  • FSDetView(TPAMI2022) Few-Shot Object Detection and Viewpoint Estimation for Objects in the Wild [paper link][arxiv link][project link][tested on datasets ObjectNet3D, Pascal3D+ and Pix3D][the first to conduct this joint task of object detection and viewpoint estimation in the few-shot regime.]

Semi-Supervised Learning

  • SSV(CVPR2020) Self-Supervised Viewpoint Learning From Image Collections [paper link][arxiv link][code|official][Unsupervised Learning, Head Pose Estimation, trained on 300W-LP and tested on BIWI, NVlabs]

  • ViewNet(ICCV2021) ViewNet: Unsupervised Viewpoint Estimation From Conditional Generation [paper link][arxiv link][Unsupervised Learning, trained on PASCAL3D+, University of Edinburgh]

  • NVSM(NIPS2021) Neural View Synthesis and Matching for Semi-Supervised Few-Shot Learning of 3D Pose [paper link][arxiv link][code|official][Johns Hopkins University, trained on PASCAL3D+ and KITTI]

  • 👍FisherMatch(CVPR2022 Oral) FisherMatch: Semi-Supervised Rotation Regression via Entropy-Based Filtering [paper link][arxiv link][project link][code|official][PKU, the first author Yingda Yin 尹英达][The visualizaion of matrix Fisher distribution so3_distribution_visualization][3DoF rotation estimation, based on FixMatch and Semi_Human_Pose, maybe suitable for 3D head pose estimation, the Semi-Supervised Rotation Regression task]

  • UCVME(AAAI2023) Semi-Supervised Deep Regression with Uncertainty Consistency and Variational Model Ensembling via Bayesian Neural Networks [paper link][arxiv link][code | official][Semi-Supervised Rotation Regression]

  • FisherMatch+(arxiv2023.05)(submitted to TPAMI) Towards Robust Probabilistic Modeling on SO(3) via Rotation Laplace Distribution [arxiv link][It proposed a new robust probabilistic modeling method; It is an extended version of FisherMatch]

▶for 3D Reconstruction

  • 3d-recon(ECCV2018) Learning Single-View 3D Reconstruction with Limited Pose Supervision [paper link][pdf link][code|official][Guandao Yang (杨关道)], may still needing additional annotations such as camera pose]

  • (CVPRW2020) Semi-Supervised 3D Face Representation Learning From Unconstrained Photo Collections [paper link][Multiple Images as Inputs]

  • SSR(ICCVW2021) SSR: Semi-Supervised Soft Rasterizer for Single-View 2D to 3D Reconstruction [paper link][arxiv link][code|official][may still needing additional annotations such as silhouette]

  • SSP3D(ECCV2022) Semi-supervised Single-View 3D Reconstruction via Prototype Shape Priors [paper link][arxiv link][code|official][Zhen Xing, on datasets ShapeNet and Pix3D][based on mean-teacher and proposed two modules namely Prototype Attention Module (using multi-head self-attention) and Shape Naturalness Module (a generative adversarial learning manner)]

  • OF4HMR(TMLR2024) Using Motion Cues to Supervise Single-frame Body Pose & Shape Estimation in Low Data Regimes [openreview link][arxiv link][code|official][CVLab, EPFL + Meta AI]

  • Real3D(arxiv2024.06) Real3D: Scaling Up Large Reconstruction Models with Real-World Images [arxiv link][project link][code|official][UT Austin][The input contains only one single object instance][We scale up training data of single-view LRMs by enabling self-training on in-the-wild images][The real data for self-training involves MVImgNet and our collected real data. The data for testing involves MVImgNet, CO3D, OmniObject3D and our real data.][It is based on the TripoSR, which is the TripoSR: Fast 3D Object Reconstruction from a Single Image ]

▶for Crowd Counting

  • IRAST(ECCV2020) Semi-supervised Crowd Counting via Self-training on Surrogate Tasks [paper link][arxiv link][Sichuan University]

  • UA_crowd_counting(ICCV2021) Spatial Uncertainty-Aware Semi-Supervised Crowd Counting [paper link][arxiv link][code|official]

  • MTCP(TNNLS2023) Multi-Task Credible Pseudo-Label Learning for Semi-Supervised Crowd Counting [paper link][code|official][TJU]

  • OPT(CVPR2023) Optimal Transport Minimization: Crowd Localization on Density Maps for Semi-Supervised Counting [paper link][code|official]

  • CrowdCLIP(CVPR2023) CrowdCLIP: Unsupervised Crowd Counting via Vision-Language Model [paper link][arxiv link][code|official][HUST]

  • SSCC (ICCV2023) Calibrating Uncertainty for Semi-Supervised Crowd Counting [paper link][arxiv link][Stony Brook University, Mean-Teacher framework, A new uncertainty estimation branch]

▶for 3D Hand-Object

  • Semi-Hand-Object(CVPR2021) Semi-Supervised 3D Hand-Object Poses Estimation With Interactions in Time [paper link][arxiv link][project link][code|official][trained on HO3D dataset, UC San Diego and NVIDIA, hand pose estimation + 6-DoF object pose estimation][using the MANO hand 3DMM model]

  • S2Contact(ECCV2022) S2Contact: Graph-based Network for 3D Hand-Object Contact Estimation with Semi-Supervised Learning [paper link]][arxiv link][project link][code|official][University of Birmingham, UNIST, SUSTech]

  • THOR-Net(WACV2023) THOR-Net: End-to-End Graformer-Based Realistic Two Hands and Object Reconstruction With Self-Supervision [paper link][arxiv link][code|official][DFKI-AV Kaiserslautern + TU Kaiserslautern + NUST-SEECS Pakistan + UPM Saudi Arabia]

  • SHAR(CVPR2023) Semi-supervised Hand Appearance Recovery via Structure Disentanglement and Dual Adversarial Discrimination [paper link][arxiv link][project link][Ynagang Wang]

▶for Face Landmarks

  • (CVPR2018) Improving Landmark Localization With Semi-Supervised Learning [paper link][arxiv link][MILA-University of Montrea and NVIDIA]

  • TS3(Teacher Supervises StudentS)(ICCV2019) Teacher Supervises Students How to Learn From Partially Labeled Images for Facial Landmark Detection [paper link][Southern China University of Science and Technology and Baidu]

  • LaplaceKL(ICCV2019) Laplace Landmark Localization [paper link][arxiv link][Northeastern University, It proposes a new loss LaplaceKL; this method can be trained under the SSL setting]

  • PIPNet(IJCV2021) Pixel-in-Pixel Net: Towards Efficient Facial Landmark Detection in the Wild [paper link][arxiv link][code|official][HKUST, the first author Haibo Jin]

  • (CVPR2022) Which Images To Label for Few-Shot Medical Landmark Detection? [paper link][arxiv link][CAS, Medical Image]

  • FaceLift(CVPR2024)(arxiv2024.05) FaceLift: Semi-supervised 3D Facial Landmark Localization [paper link][arxiv link][project link][Flawless AI]