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..
Enhancing 3D Reconstruction for Dynamic Scenes
Authors: Jisang Han, Honggyu An, Jaewoo Jung, Takuya Narihira, Junyoung Seo, Kazumi Fukuda, Chaehyun Kim, Sunghwan Hong, Yuki Mitsufuji, Seungryong Kim
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental evaluations demonstrate that our proposed approach consistently achieves superior 3D reconstruction performance across various datasets featuring complex motions. |
| Researcher Affiliation | Collaboration | 1 KAIST AI 2 Sony AI 3 ETH Zürich AI Center, CVG, PRS 4 Sony Group Corporation |
| Pseudocode | No | The paper describes methodologies using text and mathematical equations, but it does not contain any explicit pseudocode or algorithm blocks. |
| Open Source Code | No | Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [No] Justification: Due to the large size of the pretrained weights, we plan to release them publicly at a later time. |
| Open Datasets | Yes | Training datasets. As shown in Table 1, we train D2USt3R on multiple datasets, including Blink Vision Outdoor [22], Blink Vision Indoor [22], Spring [29], Point Odyssey [54], and Tartan Air [45]. |
| Dataset Splits | No | The paper states: "For each epoch, we randomly sample 20,000 image pairs and the network is trained for 50 epochs... Each epoch consisted of sampling 7,750, 4,750, 2,500, 2,500, and 2,500 pairs, respectively." While sampling numbers for training are provided, there is no explicit mention of the overall training, validation, and test dataset splits in terms of percentages or absolute counts for the datasets used. |
| Hardware Specification | Yes | We train with 4 NVIDIA RTX 6000 GPUs |
| Software Dependencies | No | The paper mentions "Adam W optimizer [27]" but does not specify software environments or libraries with version numbers (e.g., Python, PyTorch, CUDA versions). |
| Experiment Setup | Yes | We freeze the encoder and fine-tune only the decoder and the DPT head [32]... For each epoch, we randomly sample 20,000 image pairs and the network is trained for 50 epochs. We use the Adam W optimizer [27] with an initial learning rate of 5e-5. ... with a batch size of 4 images per GPU and gradient accumulation steps set to 2. |