Momentum Pseudo-Labeling for Weakly Supervised Phrase Grounding

Authors: Dongdong Kuang, Richong Zhang, Zhijie Nie, Junfan Chen, Jaein Kim

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our approach achieved SOTA performance on 3 commonly used grounding datasets for weakly supervised phrase grounding tasks. Experimental results and intensive analysis on three commonly used phrase grounding datasets demonstrate the efficiency and effectiveness of our MPL approach.
Researcher Affiliation Academia 1 CCSE, School of Computer Science and Engineering, Beihang University, Beijing, China 2 School of Software, Beihang University, Beijing, China 3 Zhongguancun Laboratory, Beijing, China 4 Shen Yuan Honors College, Beihang University, Beijing, China EMAIL, EMAIL
Pseudocode No The paper describes the methodology in detail across several sections, including 'Problem Analysis' and 'Methodology', but does not present any structured pseudocode or algorithm blocks.
Open Source Code Yes Our code can be accessed via https://github.com/Kuangdd01/MPL.
Open Datasets Yes Our main experimental results are derived from benchmarks on three publicly used datasets for phase grounding task, including the Flickr30k Entities (Plummer et al. 2015), Ref COCO and Ref COCO+ (Kazemzadeh et al. 2014; Yu et al. 2016).
Dataset Splits Yes For the Ref COCO/+ dataset, we employ the UNC split (Yu et al. 2016), dividing both datasets into four parts: train, validation, test A, and test B.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments, such as GPU models, CPU types, or memory.
Software Dependencies No The paper mentions several tools and models like Faster R-CNN, Glo Ve embeddings, and transformer encoder, but it does not specify any software dependencies with their version numbers (e.g., Python, PyTorch, specific library versions) required to replicate the experiments.
Experiment Setup Yes Regarding the hyperparameter settings: the momentum update coefficient γ is set to 0.99. For FNE, the threshold ϕ is set to 0.85, and for FNC, ϕ is set to 0.95.