Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Image Stitching in Adverse Condition: A Bidirectional-Consistency Learning Framework and Benchmark
Authors: Zengxi Zhang, Junchen Ge, Zhiying Jiang, Miao Zhang, Jinyuan Liu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments show that the proposed method can generate visually pleasing stitched images under adverse conditions, outperforming state-of-the-art methods. ... Table 1: Quantitative comparison for image stitching under adverse environment. ... 5 Experiments ... 5.3 Ablation Study |
| Researcher Affiliation | Academia | Zengxi Zhang The University of Tokyo EMAIL Junchen Ge Tsinghua University EMAIL Zhiying Jiang Dalian Martime University EMAIL Miao Zhang Dalian University of Technology EMAIL Jinyuan Liu Dalian University of Technology EMAIL |
| Pseudocode | No | The paper describes the methodology through prose, mathematical equations, and diagrams (e.g., Figure 2: Workflow of the proposed Method), but it does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | Answer: [No] Justification: After the paper is accepted, we will release the code, model parameters and dataset of the proposed method. |
| Open Datasets | Yes | We further propose the first adverse scene image stitching dataset, which covers diverse parallax and scenes under low-light, haze, and underwater environments. ... We release the first real-world adverse environment image stitching dataset, which contains 2,250 degraded image pairs with homography reference to evaluate the effectiveness of stitching methods in adverse conditions. ... LSRW [48] and UIEBD [49] are used as training datasets for low-light and underwater image enhancement tasks respectively. We also synthesize the training dataset for single image dehazing from VOC [50] according to the atmospheric scattering model [51]. |
| Dataset Splits | No | LSRW [48] and UIEBD [49] are used as training datasets for low-light and underwater image enhancement tasks respectively. We also synthesize the training dataset for single image dehazing from VOC [50] according to the atmospheric scattering model [51]. ... ASIS includes 17 scenes with a total of 2,250 pairs of images, which are derived from manual capturing and network collection. |
| Hardware Specification | Yes | All the experiments are conducted on Py Torch with NVIDIA RTX 4090 GPU. |
| Software Dependencies | No | All the experiments are conducted on Py Torch with NVIDIA RTX 4090 GPU. |
| Experiment Setup | Yes | First, we pre-train the DIR module with a learning rate of 1e-4 and an epoch of 200. ... Subsequently, we trained the BCDA module with DIR with an epoch of 160. ... During the training process, the µ and σ are initialized to 0 and 32 respectively. M, N, σGT, φ are set to 8, 6, 2 and 20. Batch size and learning rate are set to 16 and 5e-5. Finally, we trained the MSC with a learning rate of 1e-4 and an epoch of 100. λ1, λ2, λ3, λ4 ,λ5 and λ6 are set to 2, 1, 100, 100, 1 and 1. |