Datasets:

---

imageomics

/

fine-grained-challenges

like 0

Follow

HDR Imageomics Institute 177

Verified the dataset mechanics against a public test copy on main (thompsonmj/test-fine-grained-challenges).

To check:

# uv pip install pylance polars pillow
import lance
import io
import polars as pl
from PIL import Image

URI = "hf://datasets/thompsonmj/test-fine-grained-challenges/data/challenges.lance"
ds = lance.dataset(URI) # remote, tokenless

ds.count_rows() # 1432

peromyscus = ds.scanner(filter="challenge_group = 'Peromyscus'",
                        columns=["uuid", "species", "emb_bioclip2"]).to_table()
# lazy scans on remote dataset
ds.scanner(filter="array_length(seen_in_training) = 0").to_table() # should be empty now; no unseen examples in this corpus
ds.scanner(filter="array_has(seen_in_training, 'bioclip-2')").to_table()
ds.count_rows(filter="species = 'Peromyscus maniculatus'") # 98

# get an image
row = ds.take([0], columns=["image", "species"]).to_pylist()[0]
Image.open(io.BytesIO(row["image"])) # webp 720x506

t = ds.to_table(columns=["challenge_group", "license_name"])
pl.from_arrow(t).group_by(["challenge_group", "license_name"]).len() # per-license counts

lance.write_dataset(peromyscus, "./peromyscus.lance") # materialize locally

This PR now prepares the complete Fine-Grained Challenges v0.1.0 release candidate.

• 6,633 images: 1,737 Peromyscus, 4,499 Ixodidae, and 397 zebra records
• 6,152 species-labeled records representing 214 species
• 305 genus-only Peromyscus camera-trap images from LILA BC, marked unseen by the listed BioCLIP model versions
• Trochilidae is not included in this release
• records with missing license links, NoDerivatives terms, all-rights-reserved labels, or ambiguous other labels are excluded
• all retained images are lossless WebP with a maximum edge of 720 pixels
• raw FP32 image embeddings are included for pinned revisions of BioCLIP, BioCLIP 2, and BioCLIP 2.5
• the clean Lance tree contains one live dataset version, four fragments, three cosine IVF-FLAT indices, and scalar indices for UUID and challenge group
• the dataset card follows the Imageomics template and records curation, provenance, training exposure, limitations, and per-image licensing

The v0.x series is explicitly pre-stable. Minor versions may change data, schema, embeddings, or scientific interpretation; patch versions are reserved for documentation and administrative corrections. The v0.1.0 tag should be created from main after this PR is merged.

This has been prepared with the assistance of gpt-5.6-sol.