OpenResearchAlpha

International Conference on Optical Signal Processing and Intelligent Algorithms (OSPIA 2026) · 12 Nov 2026

Enhancing capability boundaries of in-context learning in large language models for computer vision

Daiwen Sun, Youwei Li

0
Citations
no database records one yet
0
Citations in 2026
so far
–
Leading author's h-index
no author has a profile yet
0
References
works it cites, from open citation data

Citations by database

Each database indexes a different set of journals, preprints and books, so they are shown side by side rather than blended.

  • OpenCitations0
  • Crossref0

Citations per year

The current year is still counting.

  • 2026 so far0

Abstract

Visual context learning enables large language models to perform visual question answering and cross-scene reasoning based on a limited number of image and text examples. However, its performance is limited by factors such as example redundancy, visual token overhead, insufficient cross-modal alignment, and prediction confidence bias. To address these issues, this paper proposes a boundary-enhanced visual context learning framework, BEV-ICL. It improves the task relevance and structural diversity of examples through boundary-aware example selection and rearrangement, preserves key objects and spatial relationships using structure-aware visual token compression, achieves dynamic alignment of visual evidence with question semantics through relational conditional gating, and enhances output reliability through uncertainty-aware reasoning and probability calibration. In the constructed experimental scheme, the average accuracy of the proposed method reached 69.6%, an improvement of 4.4 percentage points compared to the comparison method. The long context accuracy increased from 57.3% to 68.9%; the number of context tokens decreased from 5120 to 3200, a reduction of 37.50%; the average inference latency decreased from 428.6 ms to 286.4 ms , a reduction of 33.18%; P95 latency decreased by 34.29%; system throughput increased from 22.4 req/s to 33.8 req/s, an increase of 50.89%; and the expected calibration error decreased from 6.8% to 4.1%. These results demonstrate that jointly optimizing example configuration, visual structure compression, cross-modal relationships, and prediction calibration can expand the applicability of visual context learning in terms of task performance, context capacity, and inference efficiency.

What the data shows

  • No database records a citation yet, as is usual for a paper this new.

Graph

This paper in the middle, its authors and their most-cited papers below, the papers it cites to the left and the papers citing it to the right. The timeline lays every paper out by year.

Citing paper shares an authorAuthor without a profileSize shows citations, or h-index for authors

Select a node for its details. Drag to pan, scroll to zoom, drag a node to move it.

Start with

Authors (2)

Each author's standing in QAI's scholar index: the works their ORCID iD is attached to, and for anyone whose ORCID record names a lab OpenResearch follows, their arXiv preprints matched by name and a shared co-author. Names are as printed on the paper.

ProfileMost-cited paper

Cited by 0 papers

Who cites this paper, from open citation data, with Crossref's record of each; the 100 newest when there are more.

0 shown
No citing papers for this filter.

References (0)

What it cites, from open citation data, with Crossref's record of each.

No references in the open citation data yet.