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..
GOATex: Geometry & Occlusion-Aware Texturing
Authors: Hyunjin Kim, Kunho Kim, Adam Lee, Wonkwang Lee
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results demonstrate that GOATEX consistently outperforms existing methods, producing seamless, high-fidelity textures across both visible and occluded surfaces. Experiments show that GOATEX produces high-quality textures across both visible and occluded regions...User studies and GPT-based evaluations show that GOATEX is strongly preferred over existing methods, achieving state-of-the-art texture quality across both visible and occluded surfaces. |
| Researcher Affiliation | Collaboration | 1KRAFTON AI 2NC AI 3UC Berkeley 4Seoul National University |
| Pseudocode | No | The method is described in sections 3.1, 3.2, and 3.3 with descriptive text and mathematical equations. For example, 'Formally, the influence weight W(f, k) of a ray r with intersection order k on face f is defined as: W(f, k) = X r Rk(f) max ( n(f) d(r), 0), (1)'. Figure 2 is an 'Overall Pipeline of GOATEX.' but not pseudocode. |
| 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: We use the Objaverse and Objaverse-XL datasets, both of which are open access. The code will not be made public at the time of submission; its release will be determined through further discussion at a later stage. |
| Open Datasets | Yes | For evaluation, we curate a diverse set of 139 assets from Objaverse [12] and 87 assets from Objaverse-XL [11]... We use the Objaverse and Objaverse-XL datasets, both of which are open access. |
| Dataset Splits | No | For evaluation, we curate a diverse set of 139 assets from Objaverse [12] and 87 assets from Objaverse-XL [11], selecting 226 high-quality meshes with detailed interior geometries to assess the performance of our method. |
| Hardware Specification | Yes | The experiment was conducted using a single RTX A6000 GPU, requiring 12GB of memory per inference. |
| Software Dependencies | No | For all experiments, we use Stable Diffusion 1.5 [40] combined with a depth-based Control Net [54] to generate multi-view images... Rendering is done with PyTorch3D [38]... Ray casting for hit level assignment was performed... using the Open3D [58] library. |
| Experiment Setup | Yes | Each view is rendered at a resolution of 768 768, with a corresponding latent resolution of 96 96. The latent UV texture map has a resolution of 512 512, and the final RGB UV texture map is generated at 1024 1024. For visibility and texture synthesis, the maximum hit level is set to 4; we define 16 hemispherical views (8 equatorial at 45 , 8 elevated at 45 ). |