Object-level Geometric Structure Preserving for Natural Image Stitching

Authors: Wenxiao Cai, Wankou Yang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experimental results demonstrate that OBJ-GSP outperforms existing methods in both pixel alignment and shape preservation. Quantitative Results. Table 1 shows MDR and NIQE results on datasets used in other stitching algorithms and our own dataset. We outperform GSP and GES-GSP in both alignment and shape preservation.
Researcher Affiliation Academia Southeast University, 2 Si Pai Lou, Nanjing JS 210096, China EMAIL
Pseudocode No The paper describes the method using mathematical equations and figures, but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes Code https://github.com/Russ Robin/OBJ-GSP
Open Datasets No We collect Stitch Bench, which is by far the largest and most diverse image stitching benchmark. Stitch Bench is currently the most comprehensive stitching test dataset. We unified the datasets from previous works and incorporated our own collected hand-held camera and aerial images to create Stitch Bench, the most complete benchmark to date.
Dataset Splits No The paper describes the composition of the Stitch Bench dataset, stating it includes "122 pairs of images from 12 works" and their own collected images, but it does not specify explicit training/validation/test splits for reproducing experiments with their method.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU models, CPU types) used for running the experiments. It mentions different backbones for SAM models (Vi T-B, Vi T-L, Vi T-H) but these are model architectures, not hardware specifications.
Software Dependencies No The paper mentions using "semantic segmentation models like the family of Segment Anything Model" and "Efficient SAM (Xiong et al. 2023)", but it does not provide specific version numbers for these or any other software libraries, frameworks, or programming languages used.
Experiment Setup Yes For fair comparison, our parameters are identical to those of GESGSP: λl = 0.75, λobj = 1.5. Our λobj corresponds λges in to GES-GSP.