davanstrien/iconclass-vlm-brillfull

Iconclass VLM — brill full labels

Training-ready VLM iconclass-classification dataset rebuilt from the fuller, cleaner source labels in biglam/brill_iconclass (CC0). Recovers labels lost to truncation in davanstrien/iconclass-vlm-sft.

• Source images: same Brill Arkyves images as biglam/brill_iconclass, bytes passed through verbatim (no re-encode).
• Labels: full Iconclass codes with operators (+n), key-combos :, and qualifiers (TEXT) kept intact. Empty/sentinel tokens stripped; ~5 malformed free-text codes dropped.
• Avg codes/image: 4.359 (vs 3.542 in the truncated iconclass-vlm-sft).
• Label cap: codes/image capped at 20 (1446 images had their long tail trimmed).
• License: CC0-1.0.

Splits

Split	Images
train	86216
test	788

The test split is a contamination-safe held-out set: images are assigned to train/test by a deterministic hash of the Brill filename, so the two image sets are guaranteed disjoint and the split is fully reproducible.

Schema (drop-in compatible)

Every row carries a superset of fields so the existing training/eval scripts work unchanged:

Column	Type	Consumer
image	Image	train_grpo.py (raw image + label)
label	list[str]	train_grpo.py ground truth
images	list[Image]	train_sft.py / eval_sft.py
messages	conversational	train_sft.py / eval_sft.py

The assistant message content is {"iconclass-codes": [...]}; the user turn uses the same instruction string as train_grpo.py.

Configs

• default — train + test splits.
• sft — train split (same rows as default/train), for train_grpo.py --dataset-config sft and train_sft.py.

Build

Built by build_brill_dataset.py. Label cap: 20 codes/image.

Research context & key finding

This dataset was built to test whether the iconclass classifier's ~25% recall ceiling was caused by truncated training labels (the original iconclass-vlm-sft was capped at 3.54 codes/image; this restores the full Brill labels at ~4.36).

Re-SFT on these fuller labels did not improve the model. Training converged well (eval_loss 0.47) but on the contamination-safe test split it scored H-F1 45.3 / hier-recall 46.4 — recall unchanged. The bottleneck is model capability (identifying the right codes), not label completeness. The lever that did work was anchored fusion (the fine-tuned model as a precision anchor + a graded VLM-judge gating in semantic-retrieval recall → H-F1 47.5 / hier-recall 57.6, with no extra training).

Splits & contamination

• train (86,216) / test (788), split deterministically by image filename hash (disjoint).
• The test split is clean for models trained on this dataset's train split. Older checkpoints trained on the overlapping iconclass-vlm-sft images are contaminated on it.