GLIMPSE: Global-Local Indexing with Multi-Precision Spectral Encoding for Explorable Video World Models

arXiv 2026 32.7 method

TLDR

GLIMPSE introduces a global-local memory interface with multi-precision spectral encoding to improve camera-controlled video world model consistency and reduce sensitivity to positional errors.

Reasoning

The paper presents a clear methodological contribution for video world models, with memory retrieval mechanisms and attenuation to handle inaccurate projections. However, the abstract lacks explicit real-world benchmarks or baselines beyond deterministic positional transport, and the claims are limited to improved geometric and photometric consistency without quantitative results visible.

Read-first score

Read-first score 32.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 34.

Recency 5%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Methodology quality 15%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=evaluation,metric

Topical relevance 25%

48.6

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 15%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code

Graph prestige 15%

5.7

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0016

Citation impact 15%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 10%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 241.

Keyword Scores

video world model

world model

interactive world model

generative world model

world simulator

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Global-local memory interface combining local patch memory for view-specific appearance and recurrent global memory for scene context
Band-wise multi-precision coordinates for addressing historical patches
Geometry-conditioned spectral attenuation to reduce sensitivity to inaccurate projected addresses
Footprint-guided retrieval that selects historical patches under a fixed local read budget

Methodology

GLIMPSE uses a global-local memory interface with local patch memory for view-specific appearance and recurrent global memory for scene context. Historical patch addresses are encoded with band-wise multi-precision coordinates and processed through geometry-conditioned spectral attenuation to reduce sensitivity to inaccurate projections. Footprint-guided retrieval selects historical patches under a fixed local read budget, and evaluation compares against deterministic positional transport under matched retrieval and read budgets using geometric correspondence and photometric consistency metrics.

Key Results

Under matched retrieval and read budgets, GLIMPSE improves geometric correspondence and photometric consistency over deterministic positional transport.

Resources

Paper