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..
Ultra-High-Definition Dynamic Multi-Exposure Image Fusion via Infinite Pixel Learning
Authors: Xingchi Chen, Zhuoran Zheng, Xuerui Li, Yuying Chen, Shu Wang, Wenqi Ren
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results show that our method maintains high-quality visual performance while fusing UHD dynamic multi-exposure images in real-time (>40fps) on a single consumer-grade GPU. Our main contributions are summarized as follows: ... We introduce 4K-DMEF, the first UHD MEF benchmark dataset for dynamic scenes. Our method, tested on both this UHD dataset and other non-UHD datasets, achieves a balance between performance and efficiency. Experiments In this section, we evaluate the proposed method by conducting comprehensive experiments on both our UHD dataset and several public non-UHD datasets. We compare our method against five state-of-the-art multi-exposure fusion (MEF) methods... In addition, we conduct ablation studies to show the effectiveness of each module within our network. |
| Researcher Affiliation | Academia | 1Shenzhen Campus of Sun Yat-sen University 2Jimei University 3The State University of New York at Buffalo 4Fuzhou University EMAIL, EMAIL, EMAIL, EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes the methodology using prose, mathematical equations (1) through (15), and block diagrams (Figure 2). There are no explicitly labeled pseudocode or algorithm blocks with structured, code-like steps. |
| Open Source Code | No | The paper does not contain any explicit statement about providing open-source code, a link to a code repository, or mention of code in supplementary materials. |
| Open Datasets | Yes | Our main contributions are summarized as follows: ... We introduce 4K-DMEF, the first UHD MEF benchmark dataset for dynamic scenes. ... Non-UHD Dataset. We also conduct experiments using two public non-UHD dynamic multi-exposure image datasets: Kalantari Dataset (Kalantari and Ramamoorthi 2017) and Mobile-HDR Dataset (Liu et al. 2023). |
| Dataset Splits | Yes | Our UHD Dynamic Multi-Exposure Image Dataset. Ultimately, we collected data for 110 UHD dynamic scenes, dividing them into 80 scenes for training and 30 for testing. ... Non-UHD Dataset. Kalantari Dataset (Kalantari and Ramamoorthi 2017) ... includes 74 training scenes and 15 test scenes. Mobile-HDR Dataset (Liu et al. 2023) ... We divide this dataset into 85 training scenes and 30 test scenes, each with a resolution of 2000 1500 pixels. |
| Hardware Specification | Yes | Implementation Details We conduct our experiments using Py Torch on a single NVIDIA GeForce RTX 4090 GPU. ... Table 1: Comparison of quantitative results on our 4K-DMEF datasets. MR denotes the maximum resolution each algorithm can handle on a single RTX 4090 GPU. |
| Software Dependencies | No | We conduct our experiments using Py Torch on a single NVIDIA GeForce RTX 4090 GPU. To optimize the network, we employ the Adam W optimizer with a learning rate 2 10 4. The paper mentions PyTorch and the AdamW optimizer but does not specify their version numbers, which is required for a reproducible description of software dependencies. |
| Experiment Setup | Yes | Implementation Details We conduct our experiments using Py Torch on a single NVIDIA GeForce RTX 4090 GPU. To optimize the network, we employ the Adam W optimizer with a learning rate 2 10 4. The network undergoes training for 1200 epochs with a batch size of 4. Additionally, the number of Feature Integration Blocks (FIBs) is 8, and the number of feature channels is 48. |