Merge ir_slo into slo modality (RAVIR is IR-SLO); 9->8 modalities

Reorganize composition by modality (complete dataset description, drop batch/date framing)

Refresh composition: add new-batch 32-cohort detail + by-modality table; fix total 1,047,909 -> 1,179,458; 57 configs

Add UWF disease config + counts -> 1,179,458 imgs / 57 cohorts

Update captions +UWF disease

Update manifest +UWF disease (131549 / 32 cohorts)

Add UWF disease dataset (7-class: AMD/DR/normal/PM/RD/RVO/uveitis) + quality/demographics

Add OCTA-500 en-face configs (+octa_enface +oct_enface) -> 1,178,758 imgs / 56 cohorts

Update captions +OCTA-500 en-face

Update manifest +OCTA-500 en-face (130,849 / 31 cohorts)

Add OCTA-500 en-face: octa_enface (vascular maps +vessel/FAZ masks) + oct_enface

Add new_public1 (6 cohorts, +slit-lamp +FA modalities) -> 1,175,758 imgs / 54 cohorts

Update new-batch manifest (127,849 imgs / 29 cohorts)

Update new-batch captions (29 cohorts)

Add new_public1: MMRDR(CFP/OCT/UWF)+MFIDDR DR, Nuclear Cataract(slit-lamp), IOVS(FA leakage)

Add FGADR config + counts -> 1,114,174 imgs / 48 cohorts

Update new-batch manifest (66,265 imgs / 23 cohorts)

Update new-batch captions (23 cohorts)

Add FGADR Seg-set (1,842 DR fundus + lesion/OD/vessel masks)

Add DeepDRiD-UWF config + DRISHTI growth -> 1,112,332 imgs / 47 cohorts

Update new-batch manifest (64,423 imgs / 22 cohorts)

Update new-batch captions (22 cohorts)

Add DeepDRiD-UWF (202) + DRISHTI labeled training (51->101)

Reconcile composition stats to actual manifest (1,112,080 imgs / 46 cohorts)

Add new-batch (21 cohorts) configs + refresh composition stats

Add new-batch images manifest

Add new-batch captions v2

Add 2026-06 new public batch (21 cohorts) image+mask shards

docs: add per-dataset image-count statistics; fix Composition totals (post quality-screening)

docs: add denoising_test config + section (separate paired OCT denoising test set)

add separate denoising/super-res TEST set (1735 paired noisy/clean OCT, 640x640) under denoising_test/

docs: note OCT quality screening (drop <256px, nyu_poag removed)

OCT quality screen: drop 58,467 low-res (<256px) imgs; remove nyu_poag(64x128), repack thoct1800/octdl

private: add 15,156 normal pseudo-labels (route-B normal-only) -> manifest+captions

docs+code: document caption v2, add disease_dict/caption_v2 pipeline

captions v2: disease/lesion-centric, replace v1

USAGE_GUIDE: add §2.6 Topcon ETDRS+demographics sidecar usage

add Topcon ETDRS sidecar (23,454 study × 74 cols: gender/DOB→age/ethnicity/axial_length/refraction/disc-fovea positions + 5-layer × 9-subfield ETDRS thickness)

add USAGE_GUIDE.md (training-side hands-on manual + new Topcon mask join)

add Topcon layer-mask supplement parquet (352,806 OCT B-scan masks, 11-class label-map PNG)

set device_technology=unknown for NYU_POAG/C8 (no assumption); add device-granularity guidance to overview

fix OCTA500 device model + add device name to all OCT captions

README: remove all-config (schema mismatch); add cast_column usage

private_topcon: complete upload (419,042 images, 255 shards, Shanghai DRI OCT Triton SS-OCT)

public_oct: complete 19-cohort upload (488,705 OCT B-scans, 210 shards)

Initial upload: public_fundus (198k images, 42 shards) + manifest + captions + code

initial commit