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..
Asymmetric Dual-Lens Video Deblurring
Authors: Zeyu Xiao, Xinchao Wang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We validate the effectiveness of As Le D-Net through extensive experiments, benchmarking it against potential solutions for asymmetric lens deblurring.4 Experiments4.1 Experimental SettingsDatasets. We use the Real MCVSR dataset [31] for our experiments. ... We present visual comparison results on the Real MCVSR dataset in Figures 6 and 7. ... Table 2 ablates the three core components. |
| Researcher Affiliation | Academia | Zeyu Xiao Xinchao Wang National University of Singapore EMAIL, EMAIL |
| Pseudocode | No | The paper describes its methodology through textual descriptions of modules (ALM, DC, RMC) and mathematical equations (Eqs. 1-6), accompanied by architectural diagrams (Figures 3-5). There are no explicit pseudocode blocks or algorithms labeled as such. |
| Open Source Code | No | The paper does not contain an explicit statement offering open-source code for the methodology described, nor does it provide a direct link to a code repository. The mention of 'publicly available code' in Section 4.2 refers to baseline methods, not the authors' own implementation. |
| Open Datasets | Yes | Datasets. We use the Real MCVSR dataset [31] for our experiments. Originally designed for multiview video super-resolution, Real MCVSR consists of triplets captured with ultra-wide, wide-angle, and telephoto lenses. |
| Dataset Splits | No | We follow the original data split of the Real MCVSR dataset. To simulate motion blur, we generate blurred frames by the widely used technique of averaging multiple consecutive frames of the video captured by different lenses [65, 48, 47, 71, 50]. We generate blurry frames for training by averaging every 7 consecutive frames, simulating varying motion blur intensities. We synthesize motion blur by averaging seven consecutive frames for both training and testing, a widely used approximation that does not fully model real-world blur. |
| Hardware Specification | Yes | Training is conducted on an NVIDIA RTX 3090 GPU. Inference settings. ... Time costs (ms) are measured on blurred frames with a resolution of 256 256 using an NVIDIA GTX 1080 Ti GPU. |
| Software Dependencies | No | The paper mentions using the Adam optimizer, Charbonnier loss, Spy Net [57] for optical flow estimation, and RAFT [68] for temporal consistency analysis. However, it does not specify version numbers for these or any other software libraries or programming languages used in the implementation. |
| Experiment Setup | Yes | We use 7 frames as input during training, with a mini-batch size of 4 and an input frame resolution of 128 128. We apply data augmentation techniques to the training data, including horizontal flips and random rotations of 90 , 180 , and 270 . As Le D-Net is trained for 300K iterations using the Adam optimizer with a Cosine Annealing learning rate scheduler. Network architecture parameters are set to N1 = 1 and N2 = 30. The number of channels is 64, and K in ALM is set to 3. Supervision is enforced using the Charbonnier loss [71] via L = qˆI IGT 2 + ε2, where ε is set to 1 10 3 in our experiments. The initial learning rate for As Le D-Net is 1 10 4. |