Paper : https://arxiv.org/abs/2307.15880 Effective Whole-body Pose Estimation with Two-stages Distillation Whole-body pose estimation localizes the human body, hand, face, and foot keypoints in an image. This task is challenging due to multi-scale body parts, fine-grained localization for low-resolution regions, and data scarcity. Meanwhile, applying a highly e arxiv.org GitHub : https://github...
이번에 읽게된 논문은 CVPR 2021에 소개된 MobileHumanPose: Toward real-time 3D human pose estimation in mobile devices 입니다. 참고로 카이스트에서 게재한 논문이며, 3D Human Pose Estimation 모델을 Mobile Device에서 작동 될 수 있도록 경량화 한 논문입니다. Contribution 본 논문에서의 주요 Contribution은 아래와 같습니다. 기존 3D HPE 방법들이 높은 computing cost + 정확도에 초점을 맞췄던 것에 비해 본 논문에서는 모바일 기반 모델 효율성을 다룸 MobileNet v2 수정, parametric activation function, Skip concatenation (U-..
GitHub : https://github.com/zhezh/occlusion_person GitHub - zhezh/occlusion_person: A dataset for multiview 3D human pose estimation with detailed occlusion labels, powered by Unr A dataset for multiview 3D human pose estimation with detailed occlusion labels, powered by UnrealCV - GitHub - zhezh/occlusion_person: A dataset for multiview 3D human pose estimation with detaile... github.com Paper ..
Paper : https://arxiv.org/pdf/2303.07399v1.pdf GitHub : https://github.com/open-mmlab/mmpose/tree/1.x/projects/rtmpose GitHub - open-mmlab/mmpose: OpenMMLab Pose Estimation Toolbox and Benchmark. OpenMMLab Pose Estimation Toolbox and Benchmark. Contribute to open-mmlab/mmpose development by creating an account on GitHub. github.com 오늘 리뷰할 논문은 Shanghai AI Lab에서 소개한 "RTMPose: Real-Time Multi-Perso..
Paper : https://www.researchgate.net/publication/368336170_Keypoint-wise_Adaptive_Loss_for_Whole-Body_Human_Pose_Estimation Introduction 오늘 읽어볼 논문은 AAAI 2023에 소개될 Keypoint-wise Adaptive Loss for Whole-Body Human Pose Estimation 라는 논문입니다. NHN Cloud 분들이 논문을 냈네요. 아직 코드는 공개가 안된 듯 합니다. 이 논문은 dense 및 coarse keypoints의 mixed characteristic을 분석하여 whole-body human pose estimation을 수행합니다. whole body를 추정하는..
humanpose estimation 연구 분야에서 접근 방식을 주로 top-down, bottom-up 방법으로 나누곤 하는데요, 이외에도 주목해야할 방식이 있습니다. 바로 one-shot approach 입니다. (CenterAttention 논문에서 나온 표현) one-shot 방법은 bottom-up 기반 방식에서 조금 더 업그레이드 된 버전이라고 보시면 되는데요, 일반적으로 각 root node의 위치를 regression 한 다음 keypoint 위치에 대한 offset을 예측하는 것입니다. 이를 pixel-wise regression이라고도 합니다. 최적화 과정이 필요한 그룹화 과정이 필요없기 때문에 훨씬 inference time이 빠릅니다. 그러나 occlusion 및 scale 변동성에..
Paper : https://arxiv.org/abs/2110.05132 The Center of Attention: Center-Keypoint Grouping via Attention for Multi-Person Pose Estimation We introduce CenterGroup, an attention-based framework to estimate human poses from a set of identity-agnostic keypoints and person center predictions in an image. Our approach uses a transformer to obtain context-aware embeddings for all detected keypoint arx..
Multi-person Pose Estimation은 아래 그림과 같이 이미지 상에 나타난 여러 사람들에 대한 keypoint 연결 관계를 추론해내는 것입니다. Multi-person Pose Estimation이 어려운 이유는 아래와 같습니다. 1. 여러명의 사람이 다양한 scale과 position에서 등장하는 문제 2. 사람들의 activity 및 interaction 으로 인해 complex pose를 유발됨 3. 여러 사람들의 body part articulation 및 occlusion로 인해 어려움 Top-down multi-people pose estimation 각 사람을 먼저 검출한 다음 검출된 이미지 패치에서 각 사람의 자세를 독립적으로 추정하는 전략입니다. 이러한 접근 방식의 장점은..
Paper : https://arxiv.org/abs/2207.02425 Self-Constrained Inference Optimization on Structural Groups for Human Pose Estimation We observe that human poses exhibit strong group-wise structural correlation and spatial coupling between keypoints due to the biological constraints of different body parts. This group-wise structural correlation can be explored to improve the accuracy an arxiv.org 이번에..
Paper : https://arxiv.org/abs/2111.08557 Rethinking Keypoint Representations: Modeling Keypoints and Poses as Objects for Multi-Person Human Pose Estimation In keypoint estimation tasks such as human pose estimation, heatmap-based regression is the dominant approach despite possessing notable drawbacks: heatmaps intrinsically suffer from quantization error and require excessive computation to ge..
Paper : https://arxiv.org/pdf/2207.10387.pdf GitHub : https://github.com/luminxu/Pose-for-Everything GitHub - luminxu/Pose-for-Everything: The official repo for ECCV'22 paper: Pose for Everything: Towards Category-Agnostic Pose E The official repo for ECCV'22 paper: Pose for Everything: Towards Category-Agnostic Pose Estimation - GitHub - luminxu/Pose-for-Everything: The official repo for ECCV'2..
Paper : https://arxiv.org/pdf/2107.10466.pdf GitHub : https://github.com/IIGROUP/PoseDet GitHub - IIGROUP/PoseDet: [FG 2021] Code for PoseDet: Fast Multi-Person Pose Estimation Using Pose Embedding [FG 2021] Code for PoseDet: Fast Multi-Person Pose Estimation Using Pose Embedding - GitHub - IIGROUP/PoseDet: [FG 2021] Code for PoseDet: Fast Multi-Person Pose Estimation Using Pose Embedding github..
기존 AdaptivePose 관련 글 https://eehoeskrap.tistory.com/664 [Paper Review] AdaptivePose : Human Parts as Adaptive Points Paper : https://arxiv.org/pdf/2112.13635.pdf 이번에 리뷰할 논문은 AAAI 2022에서 소개된 "AdaptivePose : Human Parts as Adaptive Points" 라는 논문입니다. multi-person pose estimation 방법은 보통 bottom up 이나 top down 방식으 eehoeskrap.tistory.com Paper : https://arxiv.org/pdf/2210.04014v1.pdf AdaptivePose에 이어 A..
Paper : https://arxiv.org/pdf/2112.13635.pdf Github : https://github.com/buptxyb666/AdaptivePose GitHub - buptxyb666/AdaptivePose: This is an official implementation of our AAAI2022 paper“AdaptivePose: Human Parts as Adapti This is an official implementation of our AAAI2022 paper“AdaptivePose: Human Parts as Adaptive Points” - GitHub - buptxyb666/AdaptivePose: This is an official implementation ..
Paper : https://arxiv.org/abs/2208.00090 Explicit Occlusion Reasoning for Multi-person 3D Human Pose Estimation Occlusion poses a great threat to monocular multi-person 3D human pose estimation due to large variability in terms of the shape, appearance, and position of occluders. While existing methods try to handle occlusion with pose priors/constraints, data augme arxiv.org GitHub : https://gi..
Paper : https://openaccess.thecvf.com/content/CVPR2022/papers/Kundu_Uncertainty-Aware_Adaptation_for_Self-Supervised_3D_Human_Pose_Estimation_CVPR_2022_paper.pdf 본 논문의 main contribution은 아래와 같습니다. multi-representation pose network를 사용하는 MRP-Net을 제안하였으며, pose-uncertainty는 두 가지 다양한 설계(model-free, model-based)를 기반으로 하는 2개의 output head를 통한 pose prediction간의 불일치로 정량화됩니다. 제안된 포즈와 joint uncertainty의 효율..
Paper : https://openaccess.thecvf.com/content_ICCV_2019/papers/Cheng_Occlusion-Aware_Networks_for_3D_Human_Pose_Estimation_in_Video_ICCV_2019_paper.pdf 본 논문에서는 3D Human Pose Estimation에서 Occlusion 문제를 해결하기 위하여 occlusion aware deep learning framework 제안합니다. 이를 위해 keypoint의 2D confidence heatmap과 optical flow의 consistency constraint를 사용하여 occluded keypoint의 unreliable estimation을 filtering 합니다. oc..
