Inter-image Contrastive Consistency for Multi-Person Pose Estimation
Authors: Xixia Xu, Yingguo Gao, Xingjia Pan, Ke Yan, Xiaoyu Chen, Qi Zou
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on kinds of architectures across three datasets (i.e., MS-COCO, MPII, Crowd Pose) show the proposed ICON achieves substantial improvements over baselines. Furthermore, ICON under the semi-supervised setup can obtain comparable results with the fully-supervised methods using only 30% labeled data. |
| Researcher Affiliation | Collaboration | 1 Beijing Key Laboratory of Traffic Data Analysis and Mining, Beijing Jiaotong University, Beijing, China 2 Tencent Youtu Lab, Shanghai, China |
| Pseudocode | No | The paper describes its methods through text and equations but does not include any pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain an explicit statement about the release of source code or a link to a code repository. |
| Open Datasets | Yes | Datasets and Experimental Setup MS-COCO(Lin et al. 2014)... MPII(Andriluka et al. 2014)... Crowd Pose(Li et al. 2019)... |
| Dataset Splits | Yes | SSL training set partition. We evaluate our method when different ratios of labeled instances are used on COCO and MPII. We choose 5%, 10%, 20%, 30% of the training samples as the labeled data and the remaining are unlabeled. [...] Main results on COCO val2017 and MPII val set. |
| Hardware Specification | Yes | We implement all experiments in Py Torch with 4 Tesla V100 GPUs. |
| Software Dependencies | No | The paper mentions "Py Torch" but does not specify a version number for it or any other software dependencies. |
| Experiment Setup | Yes | For COCO, the detected human boxes are resized to 256 x 192 and we trained for 210 epochs. The learning rate follows (Sun et al. 2019). For MPII, input size is 384 x 384 and the trained for 180 epochs. For Crowd Pose, the training is similar with COCO and trained for 220 epochs. For data augmentation, we apply random flip with probability of 0.5, random rotation in [ -45 ,+45 ], random resize with [0.65, 1.35] and half-body augmentations. The weighted factors in Eq. 5, 7 are set as: λ1 = 0.3, λ2 = 0.7, λp = 1, λu = 0.1. The τ = 0.07, γ = 0.75. |