Nereus connects fine-grained underwater attributes with spatially grounded counting.

A dataset and evaluation resource for asking not only what is present, but what the image supports.

Underwater scenes combine clutter, camouflage and densely overlapping organisms.

A category name does not describe morphology, condition or interaction.

A total count does not explain where organisms gather or how their surroundings relate to that pattern.

The manuscript reports 88,555 images and approximately 2.68 million question-answer pairs.

The schema spans 136 attribute dimensions. Fields are organism-specific, not universal to every object.

Heterogeneous underwater sources are organized into object-centered and point-grounded supervision.

Object crops, image context and biological priors support multi-model attribute discussion and arbitration.

Point annotations support region counts, visibility labels, density maps and evidence-linked questions.

Annotation combines model-assisted generation with targeted human review and expert calibration.

Quality ratings and selected-case arbitration are not a whole-dataset accuracy estimate.

A localized fish example links rough texture, high camouflage and a hiding-in-coral interaction label.

The cuttlefish record describes mottled skin, high camouflage and a resting posture.

The coral record combines fan-like growth, intact tissue and a coral-rock contact label.

Region-aware counting links each annotated fish to image coordinates and spatial cells.

Sample 0000 contains 90 annotated fish. Its densest grid cell is top-right, with 31.

Semantic memberships include 82 open-water and 5 reef-rock labels. Region memberships can overlap.

Camouflage-state counting separates clear, ambiguous and highly camouflaged annotations.

Sample 0001 has 134 clear, 130 ambiguous and 6 highly camouflaged point annotations.

The close-up shows source-image evidence for highly camouflaged annotations.

Colored points are annotation labels, not model predictions.

Density-hotspot reasoning starts from a localized response in a supplied annotation-derived heatmap.

In open water, the annotation relates aggregation to available swimming space and possible schooling.

Near branching coral, visible habitat structure grounds the description of the fish aggregation.

Shelter and food are annotation hypotheses, not verified causes. Depth labels are relative, not metric.

Track A uses RGB only. Track B additionally supplies a ground-truth-derived density heatmap.

Count error, relative count accuracy and grounded reasoning measure different capabilities.

Table 2 reports higher attribute accuracy after Nereus supervision for all three open backbones shown.

For InternVL2.5-8B, Table 3A reports lower count MAE and higher densest-cell accuracy under QAS.

Use the stated input conditions and source licenses. Ecological interpretations require domain validation.

Explore the data, research code and interactive examples on the Nereus project page.