Learning Disentangled Attribute Representations for Robust Pedestrian Attribute Recognition
Authors: Jian Jia, Naiyu Gao, Fei He, Xiaotang Chen, Kaiqi Huang1069-1077
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on PETA, RAPv1, PA100k, and RAPv2 show that the proposed method performs favorably against stateof-the-art methods. In this section, we first conduct experiments on four datasets PETA (Deng et al. 2014), RAPv1 (Li et al. 2016), PA100k (Liu et al. 2017), and RAPv2 (Li et al. 2018b), to make a fair comparison with state-of-the-art (SOTA) methods. The dataset information and experimental settings are given in the supplementary material. Then, we conduct exhaustive ablation studies on the largest dataset PA100k to validate the contribution of each component. |
| Researcher Affiliation | Academia | Jian Jia1, 2, Naiyu Gao1, 2, Fei He1, 2, Xiaotang Chen1, 2, Kaiqi Huang1, 2, 3 1 School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China 2 CRISE, Institute of Automation, Chinese Academy of Sciences, Beijing, China, 3 CAS Center for Excellence in Brain Science and Intelligence Technology, Shanghai, China {jiajian2018, gaonaiyu2017, hefei2018}@ia.ac.cn, {xtchen, kqhuang}@nlpr.ia.ac.cn |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide a direct link to open-source code or explicitly state that the code for the described methodology is publicly available. |
| Open Datasets | Yes | In this section, we first conduct experiments on four datasets PETA (Deng et al. 2014), RAPv1 (Li et al. 2016), PA100k (Liu et al. 2017), and RAPv2 (Li et al. 2018b), to make a fair comparison with state-of-the-art (SOTA) methods. |
| Dataset Splits | No | The paper mentions the number of training images for some datasets (e.g., '90,000 and 67,943 training images on PA100k and RAPv2 respectively') but does not explicitly provide the specific training, validation, and test dataset splits (e.g., percentages or exact counts for all splits) within the main text. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers. |
| Experiment Setup | Yes | The dataset information and experimental settings are given in the supplementary material. Then, we conduct exhaustive ablation studies on the largest dataset PA100k to validate the contribution of each component. To make an intuitive and fair comparison, all existing methods set the probability threshold 푝푡= 0.5. 퐾is a pre-defined group number and set to 퐾= 6 as default. The 퐺푚is formulated as: G푚 푘 (1 훼) G푚 푘+ 훼 G푎 푘, where 훼 (0, 1] is the momentum hyper-parameter. Experiments results on the number of the cascaded SSCA module are listed in Tab 4. As 푆increases, the performance first increases when 푆< 3 and then decreases when 푆> 3. We argue that the cascaded SSCA module is beneficial for discriminative semantic queries and accurate query attention map when 푆is small. |