CL2CM: Improving Cross-Lingual Cross-Modal Retrieval via Cross-Lingual Knowledge Transfer

Authors: Yabing Wang, Fan Wang, Jianfeng Dong, Hao Luo

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our proposed approach on two multilingual image-text datasets, Multi30K and MSCOCO, and one video-text dataset, VATEX. The results clearly demonstrate the effectiveness of our proposed method and its high potential for large-scale retrieval.
Researcher Affiliation Collaboration 1 Zhejiang Gongshang University 2 Xi an Jiaotong University 3 DAMO Academy, Alibaba Group 4 Hupan Lab, Zhejiang Province 5 Zhejiang Key Lab of E-Commerce
Pseudocode No The paper describes the methodology using mathematical equations and descriptive text, but does not include any pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not provide any explicit statement or link indicating the release of source code for the described methodology.
Open Datasets Yes We conduct experiments on two public multilingual image-text retrieval datasets (Multi30K and MSCOCO), as well as a video-text retrieval dataset (VATEX). (...) Multi30K (Elliott et al. 2016): (...) MSCOCO (Chen et al. 2015): (...) VATEX (Wang et al. 2019):
Dataset Splits Yes We adopt a similar data partition as (Young et al. 2014). (...) We follow the data split as in (Zhou et al. 2021). (...) We adopt a similar data partition as (Chen et al. 2020).
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments, such as GPU models, CPU types, or memory specifications.
Software Dependencies No The paper mentions models like CLIP and mBERT-base and an Adam optimizer, but it does not specify software dependencies with version numbers (e.g., Python, PyTorch, CUDA versions).
Experiment Setup Yes Besides, we set λ = 0.6, α = 0.4, and τ = 0.07 in our experiments. The batch size is 128, and an Adam optimizer with an initial learning rate 2.5e-5 and adjustment schedule similar to (Luo et al. 2022) is utilized.