PaperScope
LIVE · 2026-09-11 05:40 UTC

Multimodal Taxonomic Conditioning for Generative Plankton Imagery

Daniela Ivanova, Ozgu Goksu, Nicolas Pugeault

Latestcs.CLcs.LGcs.AIcs.CV
arXiv ID
2609.11673 v1
Category
Submitted
2026-09-10

Abstract

Automated plankton imaging produces severely long-tailed datasets, where the rare taxa of greatest ecological interest have too few images to train or evaluate classifiers reliably. We generate synthetic plankton imagery conditioned on taxonomy: a CLIP encoder is adapted on a large plankton corpus with a ranked contrastive objective extended to deep, ragged taxonomies, then frozen to condition a parameter-efficient diffusion transformer. We evaluate synthetic sample quality on distributional fidelity and downstream classifier utility.

Comment: European Conference on Computer Vision (ECCV) 2nd Workshop on Marine Vision

arXiv abs page · PDF · same-day batch