Wenbin Li

—

Unclaimed · AI-structured profileOpenAlex

Computer vision and LLM-agent researcher building gigapixel UAV perception benchmarks and agentic environments from scientific code

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-09-19) 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—

Not yet scored: this Evolve snapshot has no publication-volume data for Wenbin Li. The output axis would measure the number and cadence of papers, authorship roles, and venue spread. Only two 2026 arXiv preprints are on record, both in middle-author positions.

Impact—

Not yet scored: citation counts, h-index, and venue-level influence are absent from the input data, so this axis cannot be reported. The impact axis would measure how often the two 2026 preprints are cited and taken up by later agent-environment and UAV benchmark work.

Focus0/100

Scored 0 for topical concentration: the field distribution splits evenly (0.5 computer vision, 0.5 machine learning / AI for science on agent environments and LLM training infrastructure). This reflects a genuine dual-track pattern across only two cataloged papers; a longer record would show whether the split stabilizes.

Collaboration78/100

Scored 78 from real graph data: about 50 distinct coauthors appear across the cataloged work, with twelve frequent collaborators, including Yuhang Zhu, Meiyi Zhu, Yunkai Dang, Zhangnan Li, Yuxuan Wang, Hongbing Pan, Hejia Geng, Zesen Huang, Haoyang Li, Koutian Wu, Zihan Zhou, and Yuanbo Pang. This indicates a strongly networked, team-based working style.

Activity—

Not yet scored: recent publication velocity and year-over-year activity are unavailable in this snapshot. The activity axis would measure output per year and submission cadence. The only dated signal is that both cataloged preprints appeared in 2026.

Reproducibility55/100

Scored 55, the mean across two reviewed papers both rated medium reproducibility readiness. ScienceIDE shows real eval, SFT, and RL gains but with selective benchmark reporting; the gigapixel UAV benchmark is a heavy 122-megapixel artifact whose VQA was co-built by an evaluated model and whose streaming evaluation is Car-only.

Research summary

Wenbin Li is an early-stage researcher whose Evolve-reviewed work spans two equally weighted lines: computer vision (0.5 of the field distribution) and machine learning / AI for science focused on agent environments and LLM training infrastructure (0.5). The first cataloged line contributes ScienceIDE, which turns the world's scientific codebase into calibrated agent-learnable environments; Evolve's verdict credits real evaluation, supervised fine-tuning, and reinforcement learning gains with candid reporting, while noting narrow coverage and selectively reported benchmark results. The second line contributes a hard gigapixel-scale UAV benchmark for wide-area spatio-temporal perception, introducing a 122-megapixel temporal benchmark with an s-HOTA latency metric; Evolve flags that its VQA was co-built by the evaluated GPT-5.5 and that streaming evaluation covers the Car category only. Collaboration is the strongest scored axis at 78, grounded in about 50 distinct coauthors and a recurring core of twelve frequent collaborators. Reproducibility scores 55, a mean across two papers both rated medium readiness. No output, impact, or activity scores exist yet, and both cataloged papers are 2026 arXiv preprints in middle, non-corresponding author positions, so this profile should be read as an early snapshot.

Collaboration pattern

Wenbin Li works in sizeable teams: the collaboration dimension scores 78 on roughly 50 distinct coauthors across only two cataloged papers, with twelve frequent collaborators — Yuhang Zhu, Meiyi Zhu, Yunkai Dang, Zhangnan Li, Yuxuan Wang, Hongbing Pan, Hejia Geng, Zesen Huang, Haoyang Li, Koutian Wu, Zihan Zhou, and Yuanbo Pang — recurring across this network. Both papers sit in middle, non-corresponding positions, consistent with contribution inside larger group efforts.

Research keywords

LLM agentsScientific codeVerifiable rewardsReinforcement learning environmentsSupervised fine-tuningCode repairUAV perceptionGigapixel imageryObject detectionMulti-object trackingStreaming perceptionBenchmark evaluation

Highlights

ScienceIDE: scientific codebases become agent-learnable environments
Evolve's review credits ScienceIDE with converting scientific code into calibrated agent environments and reporting real evaluation, SFT, and reinforcement learning gains, while cautioning that coverage is narrow and benchmark gains are selectively reported.
A hard 122-megapixel benchmark for wide-area UAV perception
The gigapixel dynamic-scene paper is rated HARD by Evolve: it contributes a 122-megapixel temporal UAV benchmark with an s-HOTA latency metric, though its VQA was co-built by the evaluated GPT-5.5 and streaming evaluation covers the Car category only.
A wide collaboration web around two 2026 releases
The scored collaboration dimension (78) rests on roughly 50 distinct coauthors, with a recurring core of twelve frequent partners spanning both the computer-vision and agent-infrastructure lines of work.

Timeline

2026Published the ScienceIDE preprint on arXiv, turning the world's scientific codebase into calibrated agent-learnable environments with evaluation, SFT, and reinforcement learning gains.
2026Released the gigapixel dynamic-scene study on arXiv, introducing a 122-megapixel temporal UAV benchmark for wide-area spatio-temporal perception with an s-HOTA latency metric.
2026Both Evolve-reviewed papers landed in the same year, splitting the field distribution evenly between computer vision and AI-for-science agent infrastructure, alongside roughly 50 distinct coauthors.

Papers in the Evolve catalog

Frequent co-authors

Yuhang Zhu×1 Meiyi Zhu×1 Yunkai Dang×1 Zhangnan Li×1 Yuxuan Wang×1 Hongbing Pan×1 Hejia Geng×1 Zesen Huang×1 Haoyang Li×1 Koutian Wu×1 Zihan Zhou×1 Yuanbo Pang×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.

Bibliometric scores are derived from OpenAlex-based public bibliographic data only; no citation counts, h-index, or venue-level impact are available.
Output, impact, and activity axes are unscored (dimMissing); the focus (0) and reproducibility (55) scores rest on only two papers, both rated medium reproducibility readiness.
The cataloged timeline contains only 2026; no earlier career history is derivable from the provided data, so this is an early-stage profile.
No institution, country, or ORCID is provided; "Wenbin Li" is a common name, so this profile is name-resolved from public bibliographic data and both papers are middle-author, non-corresponding positions.