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..

Under the Shadow: Exploiting Opacity Variation for Fine-grained Shadow Detection

Authors: Xiaotian Qiao, Ke Xu, Xianglong Yang, Ruijie Dong, Xiaofang Xia, Jiangtao Cui

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate that our method outperforms the baselines qualitatively and quantitatively, enhancing a wide range of applications, including shadow removal, shadow editing, and 3D reconstruction.
Researcher Affiliation Academia 1School of Computer Science and Technology, Xidian University, China 2Guangzhou Institute of Technology, Xidian University, China
Pseudocode No The paper describes methods using mathematical equations and block diagrams (Figure 2), but does not contain explicitly labeled pseudocode or algorithm blocks.
Open Source Code No We will release the dataset and code to the community.
Open Datasets Yes We conduct experiments on both the proposed FSD dataset and the existing ISTD [43] dataset, which contains 1,330 training images and 540 test images. Furthermore, we utilize two shadow removal datasets, i.e., ISTD+ [26] and SRD [35], for downstream applications. ... To validate the effectiveness of the opacity shadow mask, we conducted supplementary experiments on the LRSS soft shadow dataset [10], which contains 46 pairs of shadow and shadow-free images.
Dataset Splits Yes We conduct experiments on both the proposed FSD dataset and the existing ISTD [43] dataset, which contains 1,330 training images and 540 test images.
Hardware Specification Yes All experiments are conducted on a single NVIDIA RTX 4090 GPU.
Software Dependencies No We use Efficient Net-B3 [37] as the backbone network and Image Net [5] for weights initialization. Our model is trained using the Adam optimizer for 20 epochs.
Experiment Setup Yes Our model is trained using the Adam optimizer for 20 epochs, with an initial learning rate of 5e-4, dynamically adjusted through an exponential decay strategy (decay rate of 0.7). During training, the input images are resized to 400 400 with a batch size of 12. The random horizontal flipping operation is further used for data augmentation. The weight parameters λpos, λopa, λarea, λgrad, and λrecon are set to 1, 1, 10, 10, and 10, respectively.