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..
MV-CoLight: Efficient Object Compositing with Consistent Lighting and Shadow Generation
Authors: Kerui Ren, Jiayang Bai, Linning Xu, Lihan Jiang, Jiangmiao Pang, Mulin Yu, Bo Dai
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments demonstrate state-of-the-art harmonized results across standard benchmarks and our dataset, as well as casually captured real-world scenes demonstrate the framework s robustness and wide generalization. |
| Researcher Affiliation | Academia | 1Shanghai Jiao Tong University, 2Shanghai Artificial Intelligence Laboratory, 3Nanjing University, 4The Chinese University of Hong Kong, 5University of Science and Technology of China, 6The University of Hong Kong |
| Pseudocode | No | The paper describes a two-stage framework and pipeline with textual descriptions and a diagram (Figure 2), but does not contain a clearly labeled pseudocode or algorithm block. |
| Open Source Code | No | We will release the code after acceptance. |
| Open Datasets | Yes | Besides, we evaluate our method on two public benchmarks, FOSCom [53] and Objects With Lighting (OWL) [41]. |
| Dataset Splits | Yes | From the simple synthetic set, 50 scenes are randomly held out for evaluation, while the rest are used for training. |
| Hardware Specification | Yes | We train the 2D model from scratch for 15 days, and the 3D model for 3 days with 16 NVIDIA A100 (80 GB) GPUs. |
| Software Dependencies | No | The paper mentions using Adam W optimizer but does not specify version numbers for programming languages or libraries like Python, PyTorch, or CUDA. |
| Experiment Setup | Yes | We train the 2D object compositing model for 1M iterations with batch size 128 using Adam W (base lr=2e-3, weight decay=0.05, momentum parameters β1=0.9, β2=0.95), 10k iteration linear warmup followed by cosine decay to 1e-6, FP16 mixed precision, gradient clipping at 10.0, and an EMA of 0.99. For 3D object compositing model, we reduce the learning rate to 1e-3 and batch size to 32, training for 100k iterations. |