Learning Quality-Aware Representation for Multi-Person Pose Regression
Authors: Yabo Xiao, Dongdong Yu, Xiao Juan Wang, Lei Jin, Guoli Wang, Qian Zhang2822-2830
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our method outperforms previous single-stage regression-based even bottom-up methods and achieves the state-of-the-art result of 71.7 AP on MS COCO test-dev set. ... Experiments In this section, we first briefly introduce our experimental setup. Then we carry out the ablation study to investigate the effectiveness of each components of our proposed network. Finally, we conduct the comprehensive comparisons with previous state-of-the-art methods to verify the superiority of our proposed network. |
| Researcher Affiliation | Collaboration | Yabo Xiao,1,* Dongdong Yu, 2,* Xiao Juan Wang, 1, Lei Jin, 1 Guoli Wang, 3 Qian Zhang 4 1 Beijing University of Posts and Telecommunications 2 Byte Dance Inc. 3 Tsinghua University 4 Horizon Robotics |
| Pseudocode | No | The paper contains figures illustrating the network architecture and modules (e.g., "Figure 2: The schematic diagram of our proposed network..."), but no formal pseudocode or algorithm blocks in text format. |
| Open Source Code | No | The paper does not contain any explicit statement about releasing code or a link to a code repository for the described methodology. |
| Open Datasets | Yes | We conduct our experiments on widely-used pose estimation benchmark MS COCO (Lin et al. 2014), which includes 200k images with 250k human instance annotated with the positions of 17 body joints. Following previous settings, we leverage coco train2017 with 57k images for training, mini-val set with 5k images for conducting ablation studies, test-dev set with 20k images for comparing with the previous state-of-the-art methods. |
| Dataset Splits | Yes | Following previous settings, we leverage coco train2017 with 57k images for training, mini-val set with 5k images for conducting ablation studies, test-dev set with 20k images for comparing with the previous state-of-the-art methods. |
| Hardware Specification | No | The paper discusses model architectures like "HRNet-W32" and "HRNet-W48" as backbones, but does not provide any specific details about the hardware (e.g., GPU models, CPU types, memory) used to run the experiments. |
| Software Dependencies | No | The paper mentions "Adam optimizer" and uses the "MS COCO" dataset but does not provide specific version numbers for any software dependencies (e.g., Python, PyTorch, TensorFlow, CUDA versions). |
| Experiment Setup | Yes | Implementation Details. We train our proposed network via Adam optimizer with a mini-batch size of 64. The initial learning rate is set as 5e-4 and dropped to 5e-5 and 5e-6 at the 150th and 170th epochs respectively. Furthermore, the radius of center-neighboring area γ is set to 4. The loss weight of LI and LDc k are both set to 1.0. For inference, we keep the aspect ratio of raw input image and resize the short side of the images to 512 / 640 pixels. ... Each input is cropped to 512 / 640 pixels. The output size is 1/4 of the input resolution. |