Computer vision researcher focused on physics-aware video generation and embodied world models for robot learning.
2
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-21) 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—
No publication-volume metrics are available yet. This axis would measure total scholarly output — publication counts, venues, and yearly production across indexed sources. Current evidence is limited to two 2026 arXiv preprints in the Evolve catalog, so no score is given.
Impact—
No citation metrics are available. This axis would quantify scholarly influence through citation counts and uptake of the 2026 preprints; scoring is impossible until citation records are reliably linked to this author, so no estimate is made here.
Focus100/100
The focus score is 100: every paper Evolve has classified for Wei Song sits in a single field, computer vision (video generation and embodied world models). Both catalogued 2026 arXiv preprints — StrucPhysVideo and In-Context Robot Learning with VLM Agents — target physical-dynamics modeling for generative video and robot learning. The raw basis is small (one classified paper in fieldDist), so the perfectly narrow profile may partly reflect catalog coverage rather than deliberate specialization.
Collaboration64/100
The collaboration dimension scores 64, grounded in 24 distinct coauthors recorded in the bibliometric snapshot, with eleven recurring partners including Dongzhou Cheng, Taoran Yi, Ye Fang, Xingwu Zhang, Fan Feng, and Yixuan Li. On both catalogued 2026 papers Wei Song appears in a middle author position without corresponding-author status, consistent with contributing to larger team projects in compute-heavy video generation and robotics research.
Activity—
No longitudinal activity data is available yet. This axis would track research cadence — papers per year, posting intervals, and trajectory over time. Only the shared 2026 arXiv year of the two catalogued papers is known, which is insufficient to score.
Reproducibility20/100
Reproducibility scores 20, computed as the mean over one reviewed paper (reproPapers = 1). Evolve's review of StrucPhysVideo assigns low reproducibility readiness: the reported 45.5% Physics-IQ state-of-the-art result comes from single-run evaluations, and the structured-caption training corpus is a private dataset. The review records no released code, multi-seed protocols, or public data, all of which independent verification would require.
Research summary
Wei Song is a computer vision researcher whose reviewed work centers on physics-aware video generation and embodied world models. The Evolve catalog holds two 2026 arXiv preprints: StrucPhysVideo: Learning Physical Dynamics from Structured Captions and Robot Actions, classified in computer vision (video generation and embodied world models), and In-Context Robot Learning with VLM Agents. On both, Wei Song appears in a middle author position without corresponding-author status, indicating participation in larger research teams. StrucPhysVideo proposes learning action-conditioned video models from structured physics captions, using flow matching and sparse mixture-of-experts components, and claims 45.5% state-of-the-art performance on Physics-IQ; Evolve's review flags single-run evaluations and a private training dataset, assigning low reproducibility readiness (reproducibility score 20 from one reviewed paper). The bibliometric snapshot records 24 distinct coauthors and eleven recurring collaborators, yielding a collaboration score of 64, while a perfectly narrow field distribution (one classified field, focus 100) reflects a specialized profile on a still-small evidence base. Output, impact, and activity dimensions lack sufficient data and are left unscored. 'Wei Song' is a common name; this profile is name-resolved from public bibliographic data only.
Collaboration pattern
Wei Song's snapshot records 24 distinct coauthors, with eleven recurring collaborators — including Dongzhou Cheng, Taoran Yi, Ye Fang, Xingwu Zhang, Fan Feng, and Yixuan Li — supporting a collaboration score of 64. On both catalogued 2026 arXiv papers, Wei Song appears in a middle author position and is not the corresponding author, a pattern consistent with contributing specialist roles inside larger teams typical of compute-intensive video generation and robotics research.
Research keywords
Highlights
Structured physics captions for action-conditioned video models
Claimed 45.5% Physics-IQ state of the art — under scrutiny
VLM agents for in-context robot learning
A second 2026 arXiv preprint applies vision-language-model agents to in-context robot learning, extending the profile's reach from video world models into embodied agent settings; Evolve has not yet reviewed it in depth.
Timeline
- 2026StrucPhysVideo: Learning Physical Dynamics from Structured Captions and Robot Actions appears on arXiv; Evolve classifies it in computer vision (video generation and embodied world models) with low reproducibility readiness.
- 2026Evolve's review records the paper's claim of 45.5% Physics-IQ state-of-the-art performance, while flagging single-run evaluations and a private dataset for scrutiny.
- 2026In-Context Robot Learning with VLM Agents appears on arXiv, extending the catalog footprint into vision-language-model-based robot learning.
Papers in the Evolve catalog
Frequent co-authors
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.
- Radar scores are computed from public bibliometric data (OpenAlex-based) only; no ORCID, institution, or country was available to confirm identity.
- 'Wei Song' is a common name; this profile is name-resolved from public bibliographic data and may merge or split namesakes.
- Output, impact, and activity dimensions are unscored because the input contains no publication counts or citation metrics.
- Reproducibility (20) rests on a single reviewed paper; both catalogued works are arXiv preprints with no recorded peer-reviewed venue.