Lei Sun

Panasonic (Japan) · Japan

Unclaimed · AI-structured profileOpenAlex

Vision–language researcher at Panasonic (Japan) working on long-video datasets and multimodal reasoning.

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-17) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.

Research profile

Output—

Output is not yet scored: this profile lacks reliable cumulative publication totals beyond the 9 recently indexed works, so the axis would measure total and per-year production counts once the OpenAlex identity record is more complete.

Impact—

Impact cannot be scored yet: citation totals, citing works, and field-weighted influence are missing from this snapshot, and the only citation-side figures (h-index 1, i10-index 0) are too thin to rank; the axis would track citation uptake over time.

Focus0/100

Focus scores 0 in the current radar. The raw field split is actually narrow — 100% computer vision, divided evenly at 0.5 between vision-language models and video-language understanding — but OpenAlex topics scatter across five clusters (Multimodal Machine Learning Applications 9; GANs and Image Synthesis 4; Topic Modeling 3; two 2-record clusters), which pulls the concentration measure to its floor.

Collaboration58/100

Collaboration scores 58. OpenAlex logs 17 distinct coauthors in the recent window, and 12 of them recur as frequent collaborators — including Luc Van Gool, Danda Pani Paudel, Shuaicheng Niu, and Jinjin Gu. All three Evolve-catalog papers list Sun as a middle author and none as corresponding, consistent with a contributor role inside larger, team-led computer-vision projects.

Activity67/100

Activity scores 67, on an unusually concentrated record: countsByYear shows 9 works in 2026, and recent3yWorks is also 9, meaning effectively all indexed recent output landed in a single year. The burst-like pattern fits an early-stage profile, but a one-year window is too short to establish a sustained publication cadence.

Reproducibility38/100

Reproducibility averages 38 (reproMean) across 2 scored papers. KAIROS is rated medium readiness: strong leakage control over its 5,420-hour 1 FPS-annotated corpus, but model-generated annotations lack human ground-truth validation. TTL-SR is rated low: solid ablations, yet small test sets and no released code. Both verdicts pull the axis below mid-range.

Research summary

Lei Sun is a researcher affiliated with Panasonic in Japan whose public bibliometric identity, resolved from OpenAlex, centers on computer vision for vision-language and video-language understanding, with the field split evenly (0.5 / 0.5) between those two adjacent subfields. The recent record is compact and current: OpenAlex attributes 9 works to the last three years, all timestamped 2026, with a topic profile led by Multimodal Machine Learning Applications (9) and Generative Adversarial Networks and Image Synthesis (4), plus smaller clusters in topic modeling, neural network applications, and search behavior. The citation footprint is minimal (h-index 1, i10-index 0), so the output and impact axes are left unscored rather than estimated. In the Evolve catalog, Sun appears as a middle author — never corresponding — on three 2026 arXiv papers: the KAIROS long-video dataset (5,420 hours of 1 FPS-annotated video), the TTL-SR geometry-aware test-time learning method for quantitative spatial reasoning, and FRAMEWORKERS, a multi-agent framework for AI-generated video production. Evolve reviewed two of the three, and their reproducibility scores average 38 across 2 papers, reflecting limited code release and unvalidated model-generated annotations. Collaboration is broad, with 17 distinct coauthors in the recent window.

Collaboration pattern

OpenAlex records 17 distinct coauthors in the recent window, with 12 recurring collaborators, including Luc Van Gool, Danda Pani Paudel, Shuaicheng Niu, Jinjin Gu, and Gege Zhang. Across the three Evolve-catalog papers Sun is always a middle author and never corresponding, pointing to contributor roles in larger, team-led computer-vision projects; 17 distinct coauthors against only 9 recent works suggests several parallel collaboration lines rather than one stable group.

Research keywords

video-language modelinglong-form video understandingdense video annotationtemporal groundingtest-time learningspatial reasoningparameter-efficient fine-tuning (LoRA)multimodal machine learninggenerative adversarial networksimage synthesismulti-agent systemsbenchmark construction

Highlights

KAIROS: 5,420-hour long-video dataset for fine-grained video-language modeling
Evolve's review credits the dataset with strong leakage control across its 1 FPS-annotated space, time, and dynamics tracks, but flags that annotations are model-generated and lack human ground-truth validation; reproducibility readiness is rated medium.
Geometry-aware test-time learning for quantitative spatial reasoning
TTL-SR introduces geometry-consistency-driven test-time LoRA adaptation with pseudo-labeling for vision-language models; Evolve notes solid ablations but small test sets and no released code, giving low reproducibility readiness.
FRAMEWORKERS: dynamic multi-agent framework for AI-generated video production
A 2026 arXiv paper proposing a dynamic multi-agent framework for AI-generated video production; Evolve has not yet published a verdict, keywords, field label, or reproducibility rating for this work.

Timeline

2026Nine indexed works recorded in 2026 — the only year with counts in the OpenAlex countsByYear snapshot.
2026Kairos dataset paper released: 5,420 hours of 1 FPS-annotated long video for fine-grained video-language modeling.
2026TTL-SR paper posted to arXiv, introducing geometry-aware test-time learning for quantitative spatial reasoning.
2026FRAMEWORKERS paper posted to arXiv: a dynamic multi-agent framework for AI-generated video production.

Papers in the Evolve catalog

Frequent co-authors

Ruibo Ming×2 Deheng Zhang×2 Zhendong Li×2 Mengshun Hu×2 Danda Pani Paudel×2 Luc Van Gool×2 Jinjin Gu×2 Gege Zhang×1 Shuaicheng Niu×1 Gang Dai×1 Shuangping Huang×1 He Zhang×1

About this profile

This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.

Bibliometrics come from an OpenAlex-resolved identity; 'Lei Sun' is a common name, so the record may include name-ambiguity noise despite identity: openalex tagging.
Output and impact axes are unscored because reliable citation totals are unavailable in this snapshot; the reported h-index (1) and i10-index (0) are the only citation-side figures.
Only three 2026 arXiv papers from the Evolve catalog are individually reviewed; remaining indexed works are counted but not assessed.
The FRAMEWORKERS paper lacks an Evolve verdict, keywords, field label, and reproducibility rating.