Vision researcher split between zero-shot anomaly detection and neural video compression, with sustained recent output
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
14
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—
Output is not scored: the snapshot lacks a verified complete works list. This axis would measure lifetime publication volume, first-author share, and venue quality; recent yearly counts (26 works in three years) are the only available proxy.
Impact—
Impact is not scored: per-paper citation counts, citing works, and venue-level influence are unavailable in this snapshot. The axis would measure citation accumulation and uptake; only aggregate h-index 14 and i10-index 18 exist, without per-paper detail.
Focus0/100
Focus scores 0, the lowest reading, and the raw data explains why: field weight is split exactly 50/50 between computer vision (anomaly detection) and computer vision (neural video compression), while the top topics — Advanced Vision and Imaging (45), Advanced Image Processing Techniques (28), Human Pose and Action Recognition (19), Video Surveillance and Tracking Methods (18), and Image Processing Techniques and Applications (18) — span several subareas. The record reads as two parallel agendas rather than one narrow thread.
Collaboration41/100
Collaboration scores 41, a mid-to-low reading. The last three years show 26 works with only 7 distinct coauthors — a small, stable circle including Heqi Xiang, Jiayao Liu, Shanjun Zhang, Guanghao Wu, Xiaoheng Li, Yiming Chen, and Yixin Zhao. Both catalog preprints are first-authored without corresponding-author status, suggesting hands-on technical leadership within a compact team rather than a large distributed network.
Activity96/100
Activity scores 96, near the top of the scale. Annual counts climb from 5 works in 2022 to 15 in 2023, 13 in 2024, and 11 in 2025, with 2 already in 2026 — 26 works over the trailing three years. The 2023–2025 stretch is the most productive run in a record reaching back to 2005 (earlier peak: 10 works in 2016). Continued preprint output in 2026 shows the pace is ongoing.
Reproducibility55/100
Reproducibility scores 55, averaged over 2 reviewed papers. PSMP-CLIP reports ablated pixel-level gains on 7 of 11 benchmarks, but image-level results lag rivals, key hyperparameters are unreported, and no code is released. TCNeRV is judged credible — +2.2 dB over HNeRV-Boost on UVG at 3M parameters — yet validates on a single dataset and fades at the 10M scale. Both are rated medium readiness; released code and fuller implementation detail would lift this axis.
Research summary
Xuezhi Xiang is a computer vision researcher whose OpenAlex-indexed record spans two decades, with the earliest entries in 2005–2006 and a pronounced acceleration since 2023 (15 works in 2023, 13 in 2024, 11 in 2025, and 2 already in 2026; 26 works over the trailing three years). The profile carries an h-index of 14 and an i10-index of 18. Current work splits evenly — 50/50 by field weight — between computer vision for anomaly detection and neural video compression. Topic distribution is led by Advanced Vision and Imaging (45 works), Advanced Image Processing Techniques (28), Human Pose and Action Recognition (19), Video Surveillance and Tracking Methods (18), and Image Processing Techniques and Applications (18), indicating a broad imaging-and-video base rather than a single niche. Two 2026 first-author arXiv preprints anchor the catalog: PSMP-CLIP, a CLIP+SAM2 zero-shot anomaly detector with ablated pixel-level gains on 7 of 11 benchmarks, and TCNeRV, a dual-domain temporal-context implicit neural video codec reporting +2.2 dB over HNeRV-Boost on UVG at 3M parameters. Collaboration intensity is modest (score 41; 7 distinct recent coauthors), while activity is near the top of the scale (96). Reproducibility is mid-range (55, averaged over the 2 reviewed papers): PSMP-CLIP releases no code and TCNeRV is validated on a single dataset.
Collaboration pattern
Collaboration is compact and stable: 26 works over the last three years involve only 7 distinct coauthors, all recurring — Heqi Xiang, Jiayao Liu, Shanjun Zhang, Guanghao Wu, Xiaoheng Li, Yiming Chen, and Yixin Zhao. The collaboration score of 41 reflects this small footprint. Both reviewed preprints are first-authored without corresponding designation, consistent with a hands-on contributor inside a fixed team rather than a hub of large multi-group projects.
Research keywords
Highlights
PSMP-CLIP: patch-prompt SAM and multi-semantic prompting for zero-shot anomaly detection
TCNeRV: dual-domain temporal context for implicit neural video compression
Timeline
- 2005Earliest indexed works appear in the OpenAlex record (1 work in 2005), with sparse output through the following decade.
- 2016A first local peak: 10 indexed works in 2016, the highest annual count before 2023.
- 2023Output accelerates: 15 indexed works in 2023, the record's annual maximum, followed by 13 in 2024 and 11 in 2025.
- 2026Two first-author arXiv preprints enter the catalog: PSMP-CLIP in zero-shot anomaly detection and TCNeRV in implicit neural video compression.
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.
- Bibliometrics come from an OpenAlex name-resolved record (identity: openalex) tied to ORCID 0000-0002-6185-833X; institution and country are absent from the source data.
- Output and impact dimensions are unscored: no verified complete works list or per-paper citation detail is available in the snapshot.
- Only two papers (both 2026 arXiv preprints) carry full Evolve reviews; the reproducibility figure rests on just those 2 papers.
- The name is not unique in bibliographic databases; despite ORCID anchoring, the record may still mix in work by other authors with the same name.