Heqi Xiang

—

Unclaimed · AI-structured profileOpenAlex

Early-stage computer vision researcher in zero-shot anomaly detection and implicit neural video compression

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—

Insufficient data: this axis would measure cumulative publication volume and venue mix, but the snapshot resolves only two 2026 arXiv preprints (recent3yWorks: 2) with no fuller publication list, so output cannot be scored yet.

Impact—

Insufficient data: this axis would capture citations, h-index and influence on follow-up work; the snapshot reports h-index 0 and i10-index 0, which we treat as an incomplete citation record rather than evidence of zero impact.

Focus0/100

Focus scores 0 because the record splits evenly between two distinct problems — computer vision anomaly detection (0.5) and neural video compression (0.5) — and five OpenAlex topics, from Domain Adaptation and Few-Shot Learning to Human Pose and Action Recognition, each appear only once. The two 2026 preprints support a deliberate two-track, exploratory portfolio rather than one concentrated line.

Collaboration41/100

Collaboration scores 41. OpenAlex resolves 7 distinct coauthors — Xuezhi Xiang, Jiayao Liu, Shanjun Zhang, Guanghao Wu, Xiaoheng Li, Yiming Chen and Yixin Zhao — and on both reviewed 2026 papers Heqi Xiang holds a middle author position without corresponding status, indicating contribution within a team rather than leadership of an independent group so far.

Activity32/100

Activity scores 32. The recent three-year window contains 2 works, both 2026 arXiv preprints — the zero-shot anomaly detection and the neural video compression papers — so current output is real but thin, and the countsByYear signal covers only 2026 (2 papers), leaving the longer-term publishing cadence unmeasurable.

Reproducibility55/100

Reproducibility scores 55 (reproMean over 2 assessed papers), both rated medium readiness. The reviews name concrete gaps: PSMP-CLIP ships no code and omits key hyperparameters, while TCNeRV is validated on a single dataset (UVG) and its gain fades at 10M parameters — reproducible in principle, but dependent on author clarification.

Research summary

Heqi Xiang is an early-stage computer vision researcher whose visible record, resolved from OpenAlex bibliographic data, centers on two complementary threads: zero-shot anomaly detection with vision-language models, and implicit neural representations for video compression. Both known works appeared as arXiv preprints in 2026, on each of which Xiang holds a middle author position rather than the corresponding role. The first, PSMP-CLIP, combines patch-prompted SAM2 with multi-semantic CLIP prompting for zero-shot anomaly detection and segmentation; Evolve's review credits ablated pixel-level gains on 7 of 11 benchmarks while noting that image-level performance lags rivals and key hyperparameters are unreported. The second, TCNeRV, introduces dual-domain temporal context modeling into implicit neural video compression, improving on HNeRV-Boost by 2.2 dB on UVG at roughly 3M parameters, though the benefit is single-dataset and fades at 10M-parameter scale. Bibliometrically, the profile is thin: 2 works in the last three years, 7 distinct coauthors, an even 0.5/0.5 split between anomaly detection and neural video compression, and five OpenAlex topics each observed once — indicating an exploratory, two-track agenda. Citation-based output and impact cannot be scored yet. Reproducibility averages 55 across the two assessed papers, both medium readiness, mainly due to missing code, unreported hyperparameters, and limited benchmark coverage.

Collaboration pattern

The record resolves 7 distinct coauthors — Xuezhi Xiang, Jiayao Liu, Shanjun Zhang, Guanghao Wu, Xiaoheng Li, Yiming Chen, Yixin Zhao — but on both reviewed 2026 papers Heqi Xiang appears in a middle position and is not the corresponding author, indicating team-based work without claimed leadership so far. Whether these collaborators form one stable group or several ad-hoc teams cannot be determined from the current snapshot.

Research keywords

zero-shot anomaly detectionvision-language modelsprompt learningsegment anything model (SAM)anomaly segmentationimplicit neural representationneural video codectemporal context modelingfeature fusionresidual coding

Highlights

Zero-shot anomaly detection with patch-prompt SAM2 and multi-semantic CLIP prompting
PSMP-CLIP pairs patch-prompted SAM2 with multi-semantic CLIP prompting; Evolve's review credits ablated pixel-level gains on 7 of 11 benchmarks, while noting image-level performance lags rivals, key hyperparameters are unreported, and no code is released (medium reproducibility readiness).
Dual-domain temporal context modeling for implicit neural video compression
TCNeRV injects dual-domain temporal context into INR-based video coding, reported at +2.2 dB over HNeRV-Boost on UVG with roughly 3M parameters; the verdict calls the result credible but single-dataset, with the advantage fading at 10M-parameter scale.
An exploratory two-track early portfolio
Field distribution splits evenly between anomaly detection (0.5) and neural video compression (0.5), five OpenAlex topics each appear once, and only 2 works fall in the recent three-year window with 7 distinct coauthors — breadth-first exploration rather than a settled specialty.

Timeline

2026Posts PSMP-CLIP on arXiv: patch-prompted SAM2 plus multi-semantic CLIP prompting for zero-shot anomaly detection, with ablated pixel-level gains reported on 7 of 11 benchmarks.
2026Posts TCNeRV on arXiv: dual-domain temporal context modeling for implicit neural video compression, reported at +2.2 dB over HNeRV-Boost on UVG with about 3M parameters.
2026Ends the visible window with 2 works in the recent three-year period and 7 distinct coauthors, splitting output evenly between anomaly detection and neural video compression (0.5/0.5).

Papers in the Evolve catalog

Frequent co-authors

Xuezhi Xiang×2 Jiayao Liu×2 Shanjun Zhang×2 Guanghao Wu×1 Xiaoheng Li×1 Yiming Chen×1 Yixin Zhao×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 OpenAlex only; no institution, country, or ORCID was available, so this profile is name-resolved from public bibliographic records.
Citation-based output and impact are flagged insufficient by the platform and are not scored; the recorded h-index and i10-index of 0 reflect a sparse record, not verified zero impact.
Coverage is limited to the two 2026 arXiv preprints reviewed by Evolve; no peer-reviewed venue, award, or career-history data is included.
The reproducibility mean of 55 is based on only 2 assessed papers and may shift as more work is reviewed.