sg-fish-id-bioclip-v3-test

CaesiumSG/sg-fish-id-bioclip-v3-test

sg-fish-id-bioclip-v3-test is a machine learning model from CaesiumSG. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.

BioCLIP v2 (ViT-L/14) fine-tuned on 90,410 images spanning 993 Singapore fish species.

Downloads · 30 days

Access

Public

Updated Jul 6, 2026

Repo size

37.7 MB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.ts37.7 MB · 100%

At a glance

License: mit
Access: Public
Created: Jul 6, 2026
Updated: Jul 6, 2026
SHA: 58ebeb42

Model card

From the Hugging Face model README

SGFishCV — Singapore Fish Species Identifier

BioCLIP v2 (ViT-L/14) fine-tuned on 90,410 images spanning 993 Singapore fish species.

About this model

Attribute	Value
Base model	`imageomics/bioclip-2` (ViT-L/14)
Training strategy	Two-stage: linear probe → partial backbone fine-tuning
Classification	Hierarchical (Family → Genus → Species)
Species	993 (152 families, 489 genera)
Training images	71,950
Rare species handling	CLIP zero-shot + genus/family fallback for 81 species with <5 train images

Usage

Upload a photo of a fish. The model returns:

Top-5 species predictions with confidence scores
Confidence tier badge: 🟢 High / 🟡 Moderate / 🟠 Low (rare species fallback)
Taxonomy info: Family, Genus, Order for each prediction

Confidence Tiers

Badge	Meaning
🟢	Top-1 hybrid score ≥ 60%. Identification is reliable.
🟡	Score 30–60%. Likely correct — check top-5 alternatives.
🟠	Score < 30% OR species has <5 training images. Genus/family shown instead.

Dataset

CaesiumSG/singapore-fish-species-image-dataset-split

Model Repository

The model weights, cached text embeddings, and label mappings are stored in the companion Model Repository: CaesiumSG/sg-fish-bioclip-v2.

Set the HF_MODEL_REPO environment variable if you fork this Space to point to a different model repo.

License: mit
Created: Jul 6, 2026
Updated: Jul 6, 2026