Lars Petersson

Commonwealth Scientific and Industrial Research Organisation · Australia

Unclaimed · AI-structured profileOpenAlex

CSIRO computer-vision researcher spanning uncertainty-aware 3D reconstruction and ML for biodiversity monitoring

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—

No reliable output metrics are available: the OpenAlex record shows h-index 0 and i10-index 0, so this axis is left unscored. Output would measure the volume and consistency of publications over time — total works, yearly cadence, and venue spread.

Impact—

Impact data is insufficient: citation counts in the source record are effectively absent (h-index and i10-index both 0), so no influence score is computed. This axis would measure citations, citing works, and downstream uptake of the two reviewed preprints.

Focus0/100

Focus scores 0, reflecting breadth rather than concentration in the current record: field distribution splits exactly 50/50 between computer vision (3D reconstruction and uncertainty estimation) and computer vision/machine learning for biodiversity. Topic counts reinforce the spread — advanced neural network applications, domain adaptation and few-shot learning, and multimodal machine learning each carry 4 mentions, with single counts for genomics/phylogenetics and emotion recognition.

Collaboration32/100

Collaboration scores 32, based on a recent three-year window with 6 works and only 4 distinct co-authors. The recurring names — Mohammad Ali Armin, David Ahmedt-Aristizabal, and Russell Tsuchida — indicate a tight, repeat CSIRO-based group rather than a wide external network. On both cataloged 2026 preprints Petersson sits in the last author position (not corresponding), consistent with a senior supervisory role within a small stable team.

Activity57/100

Activity scores 57 on thin evidence: the record lists 6 works in the recent three-year window, and countsByYear attributes all 6 to 2026, with no earlier years present. Read literally, 2026 is an intense year with two arXiv preprints already cataloged and reviewed, but a single-year signal may reflect when works were indexed rather than a sustained rate. Treat it as a snapshot.

Reproducibility55/100

Reproducibility scores 55, computed as the mean readiness rating across the 2 papers Evolve has reviewed. Both sit at medium readiness: GeoCond is described as well-evaluated and honest about limits, though gains are modest, while InfoTaxa is judged careful but single-dataset with approximate figure-read baselines. The consistent middling rating suggests solid evaluation practice without full artifact-level confidence.

Research summary

Lars Petersson is a computer-vision researcher at Australia's Commonwealth Scientific and Industrial Research Organisation (CSIRO), profiled here from OpenAlex bibliographic data. The available record splits evenly between two strands: computer vision for 3D reconstruction and uncertainty estimation, and machine learning for biodiversity. In 2026 he appears as last author on two arXiv preprints reviewed in the Evolve catalog. GeoCond adds a compact conditioning-aware reliability adapter (about 20k parameters) to a frozen VGGT feed-forward 3D reconstruction model, improving out-of-distribution AUSE from 0.32 to 0.20 and gating bundle adjustment. InfoTaxa proposes information-calibrated, label-free clustering for fine-grained visual taxonomy and audits it with probe and DNA barcoding evidence on BIOSCAN-5M with BioCLIP 2 representations. Topic profiles cluster around advanced neural network applications, domain adaptation and few-shot learning, and multimodal machine learning. The record shows 6 works in the recent three-year window with 4 distinct co-authors and a reproducibility mean of 55 across 2 reviewed papers. Citation-side metrics are absent from the current record, so output and impact are not scored.

Collaboration pattern

Lars Petersson's recent record shows a small-team pattern: 6 works in the last three years with only 4 distinct co-authors, and the same names recur — Mohammad Ali Armin, David Ahmedt-Aristizabal, and Russell Tsuchida — indicating a stable CSIRO computer-vision group. On both cataloged 2026 preprints he occupies the last author position though not the corresponding slot, consistent with senior supervision within a compact, ongoing collaboration rather than large multi-institution consortia.

Research keywords

feed-forward 3D reconstructionuncertainty estimationcamera posebundle adjustmentreliability adapterlabel-free clusteringfine-grained visual classificationbiodiversity monitoringDNA barcodingmultimodal machine learningdomain adaptation and few-shot learning

Highlights

GeoCond: a 20k-parameter reliability adapter for frozen VGGT 3D reconstruction
Last-authored 2026 arXiv preprint adding a conditioning-aware adapter that improves out-of-distribution AUSE from 0.32 to 0.20 and gates bundle adjustment; Evolve rates it well-evaluated and honest about limits, though gains are modest (reproducibility: medium).
InfoTaxa: information-calibrated label-free clustering for fine-grained taxonomy
Last-authored 2026 arXiv preprint proposing information-calibrated, label-free image clustering for fine-grained visual taxonomy, audited with probe and DNA barcoding evidence on BIOSCAN-5M with BioCLIP 2; Evolve finds species-level clustering limited by both clustering and representation quality (reproducibility: medium).

Timeline

2026Posts GeoCond, a conditioning-aware reliability adapter for feed-forward 3D reconstruction on frozen VGGT, as a last-authored arXiv preprint.
2026Posts InfoTaxa, information-calibrated label-free clustering for fine-grained visual taxonomy with probe and DNA barcoding audit, as a last-authored arXiv preprint.
2026Both 2026 preprints enter the Evolve review catalog, each rated medium reproducibility readiness with a combined mean of 55 across 2 papers.

Papers in the Evolve catalog

Frequent co-authors

Mohammad Ali Armin×2 David Ahmedt-Aristizabal×1 Russell Tsuchida×1 David Ahmedt-Aristizabal×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 record (identity: openalex) with no ORCID link; that record lists h-index 0 and i10-index 0, so output and impact axes are left unscored rather than estimated.
Only 2 papers from the Evolve catalog underlie the review-based judgments; the reproducibility score is a medium-confidence mean (55) over those 2 papers.
countsByYear contains a single active year (2026, 6 works), so activity readings cover a very short window and cannot indicate a longer-term trend.
The 3-year window counts only 4 distinct co-authors; collaboration patterns beyond Evolve's catalog are not captured in this data.