SPLATIFY: Reproduce, Discover, Innovate! From Papers and Ideas to Trainable 3DGS Code
TLDR
SPLATIFY is a multi-agent framework converting 3D Gaussian Splatting papers into trainable gsplat implementations, using grammar-constrained synthesis, visual feedback, and a 30-paper benchmark.
Reasoning
The paper presents a novel framework with clearly described technical innovations and a benchmark for evaluation. However, limited visible abstract details mean the full experimental outcomes are not evident, and the scope is specific to 3DGS rather than general research automation.
Read-first score
Read-first score 56.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 54.
Field roles
Rank sensitivity
Stability: volatile; rank range: 31.
Keyword Scores
Deep Analysis
Innovations
- A context-free grammar for gsplat over a modular method template with extension points for losses, densification, rendering, and optimization, constraining synthesis so generated code satisfies gsplat's architectural invariants by construction.
- Architectural elements for faithful reproduction: fork-aware citation recovery retrieving component-level code at function-level granularity, Graph-of-Thought synthesis in topological dependency order, RAG-guided in-context example selection from over 20 verified implementations, and visual feedback combining PSNR-guided regeneration, Gaussian-level structural checks, and VLM-driven patching.
- Knowledge-driven compositional improvement that autonomously finds weaknesses and composes complementary regularizers, losses, and densification strategies to improve upon original results.
- Interdisciplinary method discovery where agents retrieve physical priors from outside the 3DGS literature and compose them with rendering knowledge to produce methods for previously unaddressed scene types.
- SPLATIFY-Bench, an evaluation framework across 30 diverse 3DGS papers.
Methodology
SPLATIFY is a multi-agent framework that converts 3DGS papers into trainable gsplat-based implementations using a context-free grammar over a modular method template with extension points. It integrates fork-aware citation recovery, Graph-of-Thought synthesis, RAG-guided in-context example selection from over 20 verified implementations, and visual feedback through PSNR-guided regeneration, Gaussian-level structural checks, and VLM-driven patching. Evaluation is conducted with SPLATIFY-Bench across 30 diverse 3DGS papers, using PSNR and development time as metrics.
Key Results
On papers without public code, SPLATIFY matches expert implementations while reducing development time from weeks to minutes. Compositional discovery further improves PSNR by up to 2.4 dB, and SPLATIFY synthesizes novel methods for volumetric nebula rendering and other scientific domains.