Graph-Based Cross-Domain Knowledge Distillation for Cross-Dataset Text-to-Image Person Retrieval
Authors: Bingjun Luo, Jinpeng Wang, Zewen Wang, Junjie Zhu, Xibin Zhao
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on three publicly available text-to-image person retrieval datasets demonstrate the effectiveness of the proposed GCKD method, which consistently outperforms the state-of-the-art baselines. Experiment Setup In this section, we introduce the experiment setup of this paper, including datasets, baselines, task settings, evaluation metrics, and implementation details. Result and Analysis In this section, we present the experimental results of our method and state-of-the-art baselines on the cross-dataset text-to-image person retrieval task. We also conduct ablation studies to analyze the effectiveness of each component. |
| Researcher Affiliation | Academia | Bingjun Luo, Jinpeng Wang, Zewen Wang, Junjie Zhu*, Xibin Zhao, BNRist, KLISS, and School of Software, Tsinghua University EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes its methodology using textual descriptions, mathematical formulas (e.g., equations 1-10), and a diagram (Figure 2), but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain any explicit statement about releasing source code, nor does it provide a link to a code repository. |
| Open Datasets | Yes | In this paper, we conduct experiments on three publicly available text-to-image person retrieval datasets: ICFG-PEDES (Ding et al. 2021), RSTPReid (Zhu et al. 2021), and CUHK-PEDES (Li et al. 2017). |
| Dataset Splits | Yes | For each dataset, the image-text data is split according to the existing protocol (Bai et al. 2023; Yang et al. 2023). In the intra-dataset cross-domain experiments, we follow the cross-domain settings proposed by (Jing et al. 2020) within the CUHK-PEDES dataset. |
| Hardware Specification | Yes | All the experiments are conducted on NVIDIA Ge Force RTX 4090. |
| Software Dependencies | Yes | The proposed model is implemented based on Py Torch 1.10 framework on Python 3.8 and Ubuntu 20.04. |
| Experiment Setup | Yes | During training, the batch size is set to 4, and the optimizer is Adam W with an initial learning rate of 1e 5 and cosine scheduler strategy. The hyperparameters of the proposed method are set as follows: the number of graph layers L = 2, the number of neighbors K = 10, the temperature τ = 0.07, the momentum coefficient α = 0.999, the loss coefficients λ1 = λ2 = 0.5, λ3 = 1. |