Added metadata and readme files.

README.md

@@ -1,4 +1,160 @@
 ---
 license: cc-by-nc-4.0
+pretty_name: SynGlass
+task_categories:
+  - image-segmentation
+tags:
+  - eyeglasses-segmentation
+  - synthetic
+  - human-faces
+  - unreal-engine-5
+  - renderpeople
+  - webdataset
+  - bmvc-2026
+  - croissant
+size_categories:
+  - 100K<n<1M
+language:
+  - en
 viewer: false
 ---
+
+# SynGlass: A Large-Scale Synthetic Dataset for Fine-Grained Eyeglasses Segmentation
+
+**BMVC 2026** · Diana-Laura Borza · Alexandru-Ion Marinescu · Robert Bota · Adrian-Sergiu Dărăbant — Babeș-Bolyai University, Cluj-Napoca, Romania
+
+SynGlass is a synthetic dataset for **pixel-accurate eyeglasses lens and frame segmentation**: **332,928 renders** of **128 photorealistic RenderPeople avatars** wearing **867 distinct 3D-scanned eyeglass models**, each avatar–eyeglass pair rendered in Unreal Engine 5 under **3 random near-frontal head poses** in an indoor scene. Masks come straight from the 3D geometry, so they are free of the labelling noise that affects manually annotated thin, reflective frame contours. To our knowledge it is the largest and most diverse dataset dedicated to eyeglasses segmentation, both in image count and in number of frames.
+
+- 📄 Paper: *SynGlass: A Large-Scale Synthetic Dataset for Fine-Grained Eyeglasses Segmentation*, BMVC 2026 (link to be added)
+- 🥐 Croissant metadata: [`metadata.json`](./metadata.json) (core + Responsible-AI fields; validated with `mlcroissant`)
+- 📜 License: [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/)
+- Total size ≈ 1.3 TB (WebDataset TAR shards); start with the `test` shards (22,304 renders) if you only want to evaluate.
+
+## What is in a render ("capture session")
+
+Files are named `{subject}_{glasses}_{capture_index}_{file_type}.{ext}` and stored under the subject folder, e.g. `subject_073/subject_073_glasses_658_1_color.png`. Subject and eyeglass identifiers are anonymized (`subject_001…subject_128`, `glasses_001…glasses_867`).
+
+| Resource | Format | Content |
+|---|---|---|
+| `*_color.png` | 1024×1024 PNG (lossless) | Colour render from a frontal camera (horizontal FOV 15°, avatar at 125–175 cm) |
+| `*_mask_lens.png` | 1024×1024 RGBA PNG | **Lens mask.** Black (0,0,0,0) = not lens; any non-black pixel = lens, and its RGBA channels encode the normalized camera-view depth (see below) |
+| `*_mask_frame.png` | 1024×1024 RGBA PNG | **Frame mask**, same encoding (rim, bridge, temples) |
+| `*_meta.json` | JSON | Head rotation `head_rotation_{pitch,yaw,roll}` (deg), `head_position_*`, `eye_{left,right}_position_*`, camera positions/rotations (`{left,mid,right}_cam_*`), `camera_fov`, `slider_avatar_distance` (cm) and the rig sliders (the colour render is the frontal camera view) |
+
+Decoding a mask (NumPy):
+
+```python
+import numpy as np, imageio.v3 as iio
+rgba = iio.imread("subject_073/subject_073_glasses_658_1_mask_lens.png").astype(np.float64)  # (1024, 1024, 4), 0..255
+mask = rgba.any(axis=-1)                                     # bool: pixel belongs to the lens class
+r, g, b, a = np.moveaxis(rgba, -1, 0)
+depth = r / (256**1 - 1) + g / (256**2 - 1) + b / (256**3 - 1) + a / (256**4 - 1)   # normalized depth in (0, 1], valid where mask
+```
+
+The two lenses are separate connected components of the lens mask; the paper separates them this way and assigns predicted pixels to the nearest ground-truth lens centroid.
+
+## Eyeglass models (867) — `manifest/eyeglasses.csv`
+
+Every eyeglass model is a 3D scan of a real frame with separate frame and lens sub-meshes and is labelled along four axes:
+
+| Attribute | Values (number of models) |
+|---|---|
+| `family` (rim structure) | full-rim 839 (96.8 %) · rimless 17 (2.0 %) · semi-rimless 11 (1.3 %) |
+| `material` | plastic 726 (83.7 %) · metal 141 (16.3 %) |
+| `shape` (lens outline) | rectangle 182 · cat-eye 181 · round 153 · oval 146 · square 114 · other 47 · aviator 44 |
+| `lens_type` | clear 669 · sunglasses 198 |
+
+*full-rim*: a continuous rigid rim encloses each lens; *semi-rimless*: the rim covers only part of the boundary (usually the upper half); *rimless*: lenses attached directly to bridge and temples. The low share of rimless/semi-rimless frames reflects commercial catalogues; they are also the hardest cases for segmentation (see the paper's per-family breakdown).
+
+## Avatars (128) — `manifest/subjects.csv`
+
+Gender: 73 female · 55 male. Age group: <13: 12 · 13–18: 10 · 19–30: 73 · 31–50: 22 · 50+: 11. Ethnicity: White 73 · Black 23 · Asian 18 · Indian 6 · Middle Eastern 5 · Hispanic/Latino 3. (These are the same 128 avatars as in the authors' HumanStereo dataset.)
+
+## Head poses
+
+Each avatar–eyeglass pair is rendered three times (`capture_session_index` 1–3): two poses with yaw, pitch and roll drawn from [−2°, 2°] and one with larger deviations in [−7°, 7°] \ [−2°, 2°]. All captures are near-frontal and occlusion-free by design (cooperative subject, as in optometric measurement and frame-retrieval settings).
+
+## Splits — `manifest/sessions.csv` (column `split`)
+
+The split is **by eyeglass model**, so no frame appears in both training and test; all 128 avatars appear in both.
+
+| Split | Eyeglass models | Renders | Use |
+|---|---|---|---|
+| `train` | 776 | 297,984 | training |
+| `test` | 91 (held out: 84 full-rim, 6 rimless, 1 semi-rimless) | 22,304 | **benchmark test set** — a subsample of the test models' renders whose yaw/pitch/roll distribution matches the training split |
+| `test_pool` | the same 91 | 12,640 | the remaining renders of the test models, not part of the benchmark; **never use for training** |
+
+> ⚠️ The TAR shards under `shards/train/` were packed before the `test_pool` split was introduced and therefore physically contain the 12,640 `test_pool` renders. Always select renders through the `split` column of `manifest/sessions.csv` (version 1.1.0 of the manifest); do not train on everything found in the train shards.
+
+## Repository layout
+
+```
+README.md
+metadata.json                 Croissant 1.1 (core + RAI), record sets: splits, eyeglasses, subjects, synglass_sessions
+manifest/sessions.csv         332,928 rows: session_id, split, human_subject_name, eyeglass_model_name,
+                              capture_session_index, file_stem, color, mask_frame, mask_lens, metadata
+manifest/eyeglasses.csv       867 rows: eyeglass_model_name, split, family, material, shape, lens_type
+manifest/subjects.csv         128 rows: human_subject_name, gender, age_group, ethnicity
+shards/train/train-000000.tar … train-000142.tar   (143 shards, ≈4 GB each)
+shards/test/test-000000.tar  … test-000010.tar     (11 shards)
+shards/shards_summary.json
+```
+
+## Loading
+
+```python
+from huggingface_hub import snapshot_download
+root = snapshot_download("asergiu/SynGlass", repo_type="dataset",
+                         allow_patterns=["metadata.json", "manifest/*", "shards/test/*"],
+                         local_dir="/data/synglass")
+# extract the shards you downloaded:  for f in shards/test/*.tar; do tar -xf "$f" -C /data/synglass; done
+
+import pandas as pd
+sessions   = pd.read_csv(f"{root}/manifest/sessions.csv")
+eyeglasses = pd.read_csv(f"{root}/manifest/eyeglasses.csv")
+subjects   = pd.read_csv(f"{root}/manifest/subjects.csv")
+test = sessions[sessions.split == "test"].merge(eyeglasses, on="eyeglass_model_name", suffixes=("", "_eg"))
+row = test.iloc[0]
+color_path, lens_path = f"{root}/{row.color}", f"{root}/{row.mask_lens}"
+```
+
+The shards are WebDataset-style TARs (one sample = the four files sharing a stem), so they can also be streamed with `webdataset` without extraction.
+
+## Benchmark protocol (from the paper)
+
+- **Pre-processing:** faces are detected with MediaPipe, the box is enlarged by 20 % and re-centred on the eyes, a central crop with width = 2 × height is taken and resized to width 448.
+- **Metrics:** IoU; Boundary IoU at 1 and 3 px; and the proposed radial metrics computed per lens from the ground-truth lens centroid over 360 rays — **Radial Threshold Accuracy RTA@δ** (fraction of rays whose radial error ≤ δ px) and **Radial Error RErr** (mean absolute radial error, px). The radial formulation decomposes errors into nasal / temporal / top / bottom quadrants.
+- **Reference results (test split, IoU / BIoU@1 / RTA@1 / RErr):** UNet++ EfficientNet-B4 (CE loss) 96.36 / 65.06 / 73.84 / 1.11; SegFormer MiT-B2 96.34 / 61.68 / 70.70 / 1.09; DeepLabV3+ EfficientNet-B4 96.24 / 54.49 / 64.53 / 1.08; zero-shot SAM-2 (box prompt) 93.57 / 26.80 / 38.02 / 1.46. Full-rim frames reach BIoU@1 71.4 vs. 10.3 for rimless.
+- **Synthetic-to-real:** pre-training on SynGlass and fine-tuning improves boundary accuracy on real data (FFHQ with overlaid 3D glasses: BIoU@1 69.2 → 71.5; Roboflow lenses: IoU 80.7 → 82.5).
+
+## Intended uses and limitations
+
+**Intended:** training and benchmarking fine-grained lens/frame segmentation; pre-training for real-world eyeglasses segmentation; boundary-accuracy evaluation; research on optometric measurements (pupillary distance, fitting height, optical centre), frame retrieval and eyewear-aware face analysis.
+
+**Limitations:** entirely synthetic (indoor lighting only; real skin texture, natural illumination and reflections under uncontrolled light are not covered); near-frontal cooperative captures only (|yaw|, |pitch|, |roll| ≤ 7°); strong class imbalance across frame families (only 17 rimless and 11 semi-rimless models; the test set has 6 rimless and 1 semi-rimless model); 128 identities with the demographic skew listed above; identities are shared across splits (only eyeglass models are disjoint). A model that performs well on SynGlass may need adaptation for in-the-wild images.
+
+**Prohibited:** surveillance, covert biometric identification, or any application that could harm individuals; the license is non-commercial.
+
+## Ethics and licensing
+
+The dataset contains **no real personal data**: all faces are renders of commercially licensed RenderPeople scans, only renders are redistributed (never the 3D assets), and all subject and eyeglass identifiers are anonymized. The eyeglass models are 3D scans of commercial frames; product names are not released. No human annotators were involved. Data: **CC BY-NC 4.0**.
+
+## Maintenance
+
+Maintained by Prof. Adrian-Sergiu Dărăbant, Department of Computer Science, Babeș-Bolyai University (sergiu.darabant@ubbcluj.ro). Errata are listed here; released versions stay available with version identifiers.
+Version history: **1.0.0** (May 2026) initial release · rimless-frame renders added (Aug 2026) · **1.1.0** (Nov 2026) corrected `split` labels (`test_pool` introduced), `eyeglasses.csv` and `subjects.csv` attribute tables, Croissant RAI fields.
+
+## Citation
+
+```bibtex
+@inproceedings{borza2026synglass,
+  title     = {SynGlass: A Large-Scale Synthetic Dataset for Fine-Grained Eyeglasses Segmentation},
+  author    = {Borza, Diana-Laura and Marinescu, Alexandru-Ion and Bota, Robert and D{\u{a}}r{\u{a}}bant, Adrian-Sergiu},
+  booktitle = {Proceedings of the British Machine Vision Conference (BMVC)},
+  year      = {2026}
+}
+```
+
+## Acknowledgements
+
+Supported by the project *Romanian Hub for Artificial Intelligence – HRIA*, 2021–2027, MySMIS no. 351416, and by project PN-IV-P7-7.1-PTE-2024-0105, contract no. 43PTE (05/05/2025). 3D human scans courtesy of [RenderPeople](https://renderpeople.com/) under their commercial license.

metadata.json

@@ -4,6 +4,7 @@
     "@vocab": "https://schema.org/",
     "sc": "https://schema.org/",
     "cr": "http://mlcommons.org/croissant/",
+    "rai": "http://mlcommons.org/croissant/RAI/",
     "dct": "http://purl.org/dc/terms/",
     "citeAs": "cr:citeAs",
     "column": "cr:column",
@@ -39,58 +40,198 @@
     "regex": "cr:regex",
     "repeated": "cr:repeated",
     "replace": "cr:replace",
+    "samplingRate": "cr:samplingRate",
     "separator": "cr:separator",
     "source": "cr:source",
     "subField": "cr:subField",
-    "transform": "cr:transform"
+    "transform": "cr:transform",
+    "equivalentProperty": "cr:equivalentProperty"
   },
   "@type": "sc:Dataset",
-  "@id": "synglass-dataset",
-  "name": "SynGlass: Synthetic Eyeglasses Segmentation Dataset",
-  "description": "Synthetic dataset for fine-grained eyeglasses segmentation. It contains 332,928 mono capture sessions generated from 128 anonymized human subjects, 867 anonymized eyeglass models, and 3 captures per subject-eyeglass combination. The manifest contains 310,624 training sessions and 22,304 test sessions. Subject identifiers use the form subject_001 through subject_128, and eyeglass identifiers use the form glasses_001 through glasses_867. Each session contains a 1024x1024 color image, a frame segmentation mask, a lens segmentation mask, and a per-session metadata JSON file. The masks are lossless RGBA PNG files where black pixels indicate background/non-category pixels and non-black pixels encode normalized camera-view depth across RGBA channels. This dataset includes 6,144 rimless-eyeglasses capture sessions added to the earlier corpus.",
-  "version": "1.0.0",
+  "@id": "synglass",
+  "name": "SynGlass",
+  "alternateName": [
+    "SynGlass: A Large-Scale Synthetic Dataset for Fine-Grained Eyeglasses Segmentation",
+    "Synthetic Humans with Glasses dataset"
+  ],
+  "description": "SynGlass is a large-scale synthetic dataset for fine-grained eyeglasses segmentation, introduced at BMVC 2026. It contains 332,928 renders (1024x1024 PNG) of 128 photorealistic RenderPeople avatars wearing 867 distinct 3D-scanned eyeglass models, each combination rendered in Unreal Engine 5 under 3 random near-frontal head poses in an indoor living-room scene. Every render comes with pixel-accurate lens and frame segmentation masks generated directly from the 3D geometry (the masks additionally encode normalized camera-view depth in their RGBA channels) and a per-render JSON file with head pose, head/eye positions and camera parameters. Each eyeglass model is labelled with frame family (full-rim / semi-rimless / rimless), material (plastic / metal), lens shape (rectangle, square, round, oval, aviator, cat-eye, other) and lens type (clear / sunglasses); each avatar with gender, age group and ethnicity. The split is by eyeglass model: 776 models (297,984 renders) for training and 91 held-out models whose 22,304 pose-matched renders form the test set; the remaining 12,640 renders of the test models are kept in a separate test_pool split and must not be used for training. Subject and eyeglass identifiers are anonymized (subject_NNN, glasses_NNN). Each row of manifest/sessions.csv is one render; resource paths are relative to the dataset root and start with the subject folder.",
+  "version": "1.1.0",
   "license": "https://creativecommons.org/licenses/by-nc/4.0/",
   "url": "https://huggingface.co/datasets/asergiu/SynGlass",
+  "datePublished": "2026-05-01",
+  "dateModified": "2026-11-04",
+  "citeAs": "@inproceedings{borza2026synglass, title={SynGlass: A Large-Scale Synthetic Dataset for Fine-Grained Eyeglasses Segmentation}, author={Borza, Diana-Laura and Marinescu, Alexandru-Ion and Bota, Robert and D\\u0103r\\u0103bant, Adrian-Sergiu}, booktitle={Proceedings of the British Machine Vision Conference (BMVC)}, year={2026}}",
   "keywords": [
-    "segmentation",
-    "synthetic data",
-    "eyeglasses",
-    "computer vision",
-    "depth encoding",
-    "human avatars",
+    "eyeglasses segmentation",
+    "lens segmentation",
+    "semantic segmentation",
+    "synthetic dataset",
+    "Unreal Engine 5",
     "RenderPeople",
-    "rimless eyeglasses",
-    "frame segmentation"
+    "human avatars",
+    "eyewear",
+    "boundary accuracy",
+    "computer vision",
+    "BMVC 2026"
   ],
   "creator": [
+    {
+      "@type": "Person",
+      "name": "Diana-Laura Borza",
+      "email": "diana.borza@ubbcluj.ro",
+      "affiliation": {
+        "@type": "Organization",
+        "name": "Babeș-Bolyai University, Department of Computer Science",
+        "url": "https://www.cs.ubbcluj.ro/"
+      }
+    },
+    {
+      "@type": "Person",
+      "name": "Alexandru-Ion Marinescu",
+      "email": "alexandru.marinescu@ubbcluj.ro",
+      "affiliation": {
+        "@type": "Organization",
+        "name": "Babeș-Bolyai University, Department of Computer Science",
+        "url": "https://www.cs.ubbcluj.ro/"
+      }
+    },
+    {
+      "@type": "Person",
+      "name": "Robert Bota",
+      "email": "robertbota2201@gmail.com",
+      "affiliation": {
+        "@type": "Organization",
+        "name": "Babeș-Bolyai University, Department of Computer Science",
+        "url": "https://www.cs.ubbcluj.ro/"
+      }
+    },
+    {
+      "@type": "Person",
+      "name": "Adrian-Sergiu Dărăbant",
+      "email": "sergiu.darabant@ubbcluj.ro",
+      "affiliation": {
+        "@type": "Organization",
+        "name": "Babeș-Bolyai University, Department of Computer Science",
+        "url": "https://www.cs.ubbcluj.ro/"
+      }
+    }
+  ],
+  "publisher": {
+    "@type": "Organization",
+    "name": "Babeș-Bolyai University, Department of Computer Science",
+    "url": "https://www.cs.ubbcluj.ro/"
+  },
+  "funder": [
     {
       "@type": "Organization",
-      "name": "TVARITA Research"
+      "name": "Romanian Hub for Artificial Intelligence - HRIA (2021-2027, MySMIS no. 351416)"
+    },
+    {
+      "@type": "Organization",
+      "name": "Project PN-IV-P7-7.1-PTE-2024-0105, contract no. 43PTE (05/05/2025)"
     }
   ],
+  "conformsTo": "http://mlcommons.org/croissant/1.1",
+  "inLanguage": "en",
+  "isLiveDataset": false,
+  "rai:dataCollection": "Fully synthetic generation with an automated Unreal Engine 5 pipeline. Each of the 128 RenderPeople avatars wears each of the 867 eyeglass models (3D scans of real frames with separate frame and lens sub-meshes) and is rendered from a frontal camera (horizontal FOV 15 deg) at 125-175 cm in an indoor living-room scene, under three random head poses: two with yaw/pitch/roll in [-2, 2] deg and one in [-7, 7] deg excluding [-2, 2] deg. No real person was photographed.",
+  "rai:dataCollectionType": [
+    "Synthetic data generation (Unreal Engine 5 rendering of licensed 3D scans)"
+  ],
+  "rai:dataCollectionTimeframe": "2025 - 2026 (rimless-frame renders added in 2026)",
+  "rai:dataCollectionRawData": "Raw 1024x1024 lossless PNG renders are distributed as-is, together with the lens and frame masks rendered from the same viewpoint and the per-render JSON metadata.",
+  "rai:dataCollectionMissingData": "None: every render has a color image, a frame mask, a lens mask and a metadata file. Frames whose lenses are not visible do not occur because all captures are near-frontal and occlusion-free by construction.",
+  "rai:dataPreprocessingProtocol": [
+    "No filtering or cleaning was applied to the renders. The benchmark in the paper crops faces with MediaPipe (20 % enlarged box re-centred on the eyes, central crop with width = 2 x height, resized to width 448); the released images are the uncropped 1024x1024 renders.",
+    "The test split is a pose-matched subsample (22,304 of 34,944 renders) of the 91 held-out eyeglass models; the unselected renders are kept in the test_pool split."
+  ],
+  "rai:dataAnnotationProtocol": "Segmentation masks are rendered directly from the frame and lens sub-meshes of the 3D eyeglass models, so they are pixel-accurate and free of human labelling noise. In the RGBA mask PNGs, black means background; any non-black pixel belongs to the class and its four channels encode the normalized camera-view depth. Per-eyeglass taxonomy labels (family, material, shape, lens type) were assigned manually by the authors per eyeglass model; avatar demographics come from the RenderPeople catalogue.",
+  "rai:dataAnnotationPlatform": [
+    "None (masks from the rendering engine; taxonomy labels assigned by the authors)"
+  ],
+  "rai:dataAnnotationAnalysis": "Mask annotations are deterministic renders, so inter-annotator agreement does not apply. Taxonomy labels are a single expert assignment per eyeglass model (867 models).",
+  "rai:machineAnnotationTools": [
+    "Unreal Engine 5 (masks, depth encoding, metadata)"
+  ],
+  "rai:annotationsPerItem": "One frame mask, one lens mask and one metadata file per render.",
+  "rai:annotatorDemographics": "Not applicable (no crowd annotators; taxonomy labels assigned by the authors).",
+  "rai:personalSensitiveInformation": [
+    "The dataset contains no real personal data: all faces are renders of commercially licensed RenderPeople scans and all identifiers are anonymized (subject_NNN, glasses_NNN). Avatars carry coarse gender, age-group and ethnicity labels to enable fairness analyses. Eyeglass models are 3D scans of commercial frames; product names are not released."
+  ],
+  "rai:dataUseCases": [
+    "Training and benchmarking fine-grained eyeglasses lens/frame segmentation",
+    "Pre-training for real-world eyeglasses segmentation (synthetic-to-real transfer)",
+    "Boundary-accuracy evaluation with region-based and radial metrics (RTA, RErr)",
+    "Optometric measurement research (pupillary distance, fitting height, lens optical centre), frame retrieval, eyewear-aware face analysis"
+  ],
+  "rai:dataLimitations": [
+    "Entirely synthetic: indoor lighting only; real skin textures, natural illumination and reflections under uncontrolled light are not covered.",
+    "Near-frontal, cooperative captures only (|yaw|, |pitch|, |roll| <= 7 deg); not suited for in-the-wild or profile views.",
+    "Frame families are imbalanced as in commercial catalogues: 839 full-rim, 17 rimless, 11 semi-rimless models; the test set contains 6 rimless and 1 semi-rimless model.",
+    "128 avatars with the demographic skew listed under dataBiases.",
+    "Subject identities are shared between train and test; only eyeglass models are disjoint."
+  ],
+  "rai:dataBiases": [
+    "Avatar demographics (128): gender 73 female / 55 male; age <13: 12, 13-18: 10, 19-30: 73, 31-50: 22, 50+: 11; ethnicity White 73, Black 23, Asian 18, Indian 6, Middle Eastern 5, Hispanic/Latino 3.",
+    "Eyeglass models (867): full-rim 839 / rimless 17 / semi-rimless 11; plastic 726 / metal 141; rectangle 182, cat-eye 181, round 153, oval 146, square 114, other 47, aviator 44; clear lenses 669 / sunglasses 198."
+  ],
+  "rai:dataSocialImpact": "Intended to improve eyewear segmentation for optometric measurement, frame retrieval and eyewear-aware face analysis. Segmentation models trained on it could be used as a component of face-recognition or surveillance pipelines; the dataset is released under CC BY-NC 4.0 and must not be used for surveillance or any application that could harm individuals.",
+  "rai:dataReleaseMaintenancePlan": "Hosted on Hugging Face (asergiu/SynGlass) and maintained by Prof. Adrian-Sergiu Darabant, Babes-Bolyai University (sergiu.darabant@ubbcluj.ro). Errata are documented on the dataset page; released versions stay available with version identifiers (1.0.0: initial release; 1.1.0: corrected split labels, eyeglass taxonomy and avatar demographics tables).",
   "distribution": [
     {
       "@type": "cr:FileObject",
       "@id": "sessions-csv",
       "name": "sessions.csv",
-      "description": "One row per mono capture session. All subject and eyeglass identifiers and paths use the anonymized subject_NNN / glasses_NNN naming scheme.",
+      "description": "Manifest with one row per render: split (train / test / test_pool), anonymized subject and eyeglass ids, capture index and the relative paths of the four resources.",
       "contentUrl": "manifest/sessions.csv",
       "encodingFormat": "text/csv",
-      "sha256": "def6b0f11072f4c8099148895804aa3108b9b215119123293f76fac4cf429b32"
+      "sha256": "c0f74d8fbc8d4bfefddc785a657cdacdd8dbd6353fe789bcb10c4de4de6ce660"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "eyeglasses-csv",
+      "name": "eyeglasses.csv",
+      "description": "One row per eyeglass model (867): split, frame family, material, lens shape and lens type.",
+      "contentUrl": "manifest/eyeglasses.csv",
+      "encodingFormat": "text/csv",
+      "sha256": "497f0aa4aba0bb47f0182de790603ab39aa6d2031bad5a1c034ce3dbae8a6bd0"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "subjects-csv",
+      "name": "subjects.csv",
+      "description": "One row per avatar (128): gender, age group and ethnicity.",
+      "contentUrl": "manifest/subjects.csv",
+      "encodingFormat": "text/csv",
+      "sha256": "a11d131df40ae833d99d41ec9da9aaeb1b2e1972563561aa8bb9a51c72c2e545"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "shards",
+      "name": "WebDataset TAR shards",
+      "description": "Renders packed as WebDataset-style TAR shards, ~4 GB each: shards/train/train-NNNNNN.tar (143 files) and shards/test/test-NNNNNN.tar (11 files). Note: the train shards physically also contain the 12,640 test_pool renders; use the split column of sessions.csv to select renders.",
+      "includes": "shards/*/*.tar",
+      "encodingFormat": "application/x-tar"
     },
     {
       "@type": "cr:FileSet",
       "@id": "png-files",
       "name": "PNG images and masks",
-      "description": "Color images and segmentation masks for frames and lenses.",
+      "description": "1024x1024 colour renders (*_color.png) and RGBA frame/lens masks with depth encoding (*_mask_frame.png, *_mask_lens.png), inside the shards.",
+      "containedIn": {
+        "@id": "shards"
+      },
       "includes": "**/*.png",
       "encodingFormat": "image/png"
     },
     {
       "@type": "cr:FileSet",
       "@id": "json-files",
-      "name": "Metadata JSON files",
-      "description": "Per-session metadata files containing auxiliary capture parameters.",
+      "name": "Per-render metadata JSON",
+      "description": "*_meta.json: head rotation (pitch/yaw/roll, deg), head and eye positions, camera positions/rotations, camera_fov, slider_avatar_distance (cm).",
+      "containedIn": {
+        "@id": "shards"
+      },
       "includes": "**/*.json",
       "encodingFormat": "application/json"
     }
@@ -100,7 +241,7 @@
       "@type": "cr:RecordSet",
       "@id": "splits",
       "name": "splits",
-      "description": "Machine-learning split definitions used by the SynGlass session manifest.",
+      "description": "Split definitions used in sessions.csv. train: 776 eyeglass models, 297,984 renders; test: 91 held-out models, 22,304 pose-matched renders; test_pool: the remaining 12,640 renders of the 91 test models (not part of the benchmark, never to be used for training).",
       "dataType": "cr:Split",
       "key": {
         "@id": "splits/name"
@@ -117,8 +258,11 @@
           "@type": "cr:Field",
           "@id": "splits/url",
           "name": "url",
-          "description": "Standard Croissant semantic identifier for the split.",
-          "dataType": "cr:Split"
+          "description": "Croissant semantic identifier of the split.",
+          "dataType": [
+            "sc:Text",
+            "cr:Split"
+          ]
         }
       ],
       "data": [
@@ -129,6 +273,182 @@
         {
           "splits/name": "test",
           "splits/url": "cr:TestSplit"
+        },
+        {
+          "splits/name": "test_pool",
+          "splits/url": "cr:TestSplit"
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "eyeglasses",
+      "name": "eyeglasses",
+      "description": "Taxonomy of the 867 eyeglass models (anonymized ids).",
+      "key": {
+        "@id": "eyeglasses/eyeglass_model_name"
+      },
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "eyeglasses/eyeglass_model_name",
+          "name": "eyeglass_model_name",
+          "description": "Anonymized eyeglass model identifier (glasses_001 ... glasses_867).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "eyeglasses-csv"
+            },
+            "extract": {
+              "column": "eyeglass_model_name"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "eyeglasses/split",
+          "name": "split",
+          "description": "train (776 models) or test (91 models; their renders are in the test and test_pool splits).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "eyeglasses-csv"
+            },
+            "extract": {
+              "column": "split"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "eyeglasses/family",
+          "name": "family",
+          "description": "Frame family: full-rim, semi-rimless or rimless.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "eyeglasses-csv"
+            },
+            "extract": {
+              "column": "family"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "eyeglasses/material",
+          "name": "material",
+          "description": "Frame material: plastic or metal.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "eyeglasses-csv"
+            },
+            "extract": {
+              "column": "material"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "eyeglasses/shape",
+          "name": "shape",
+          "description": "Lens shape: rectangle, square, round, oval, aviator, cat-eye or other.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "eyeglasses-csv"
+            },
+            "extract": {
+              "column": "shape"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "eyeglasses/lens_type",
+          "name": "lens_type",
+          "description": "clear or sunglasses (tinted lenses).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "eyeglasses-csv"
+            },
+            "extract": {
+              "column": "lens_type"
+            }
+          }
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "subjects",
+      "name": "subjects",
+      "description": "Demographic attributes of the 128 RenderPeople avatars (anonymized ids).",
+      "key": {
+        "@id": "subjects/human_subject_name"
+      },
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "subjects/human_subject_name",
+          "name": "human_subject_name",
+          "description": "Anonymized avatar identifier (subject_001 ... subject_128).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "subjects-csv"
+            },
+            "extract": {
+              "column": "human_subject_name"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "subjects/gender",
+          "name": "gender",
+          "description": "female or male (as listed by RenderPeople).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "subjects-csv"
+            },
+            "extract": {
+              "column": "gender"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "subjects/age_group",
+          "name": "age_group",
+          "description": "Age group: <13, 13-18, 19-30, 31-50 or 50+.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "subjects-csv"
+            },
+            "extract": {
+              "column": "age_group"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "subjects/ethnicity",
+          "name": "ethnicity",
+          "description": "White, Black, Asian, Indian, Middle Eastern or Hispanic/Latino (as listed by RenderPeople).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "subjects-csv"
+            },
+            "extract": {
+              "column": "ethnicity"
+            }
+          }
         }
       ]
     },
@@ -136,7 +456,7 @@
       "@type": "cr:RecordSet",
       "@id": "synglass_sessions",
       "name": "synglass_sessions",
-      "description": "One record per mono capture session.",
+      "description": "One record per render (mono capture session): a colour image, a frame mask, a lens mask and a metadata file.",
       "key": {
         "@id": "synglass_sessions/session_id"
       },
@@ -145,7 +465,7 @@
           "@type": "cr:Field",
           "@id": "synglass_sessions/session_id",
           "name": "session_id",
-          "description": "Unique anonymized session identifier derived from the filename stem.",
+          "description": "Unique anonymized session identifier (= file stem).",
           "dataType": "sc:Text",
           "source": {
             "fileObject": {
@@ -160,7 +480,7 @@
           "@type": "cr:Field",
           "@id": "synglass_sessions/split",
           "name": "split",
-          "description": "Dataset split: train or test.",
+          "description": "train, test or test_pool.",
           "dataType": "sc:Text",
           "source": {
             "fileObject": {
@@ -171,14 +491,16 @@
             }
           },
           "references": {
-            "@id": "splits/name"
+            "field": {
+              "@id": "splits/name"
+            }
           }
         },
         {
           "@type": "cr:Field",
           "@id": "synglass_sessions/human_subject_name",
           "name": "human_subject_name",
-          "description": "Anonymized human subject identifier (subject_001 through subject_128).",
+          "description": "Anonymized avatar identifier; joins subjects.",
           "dataType": "sc:Text",
           "source": {
             "fileObject": {
@@ -187,13 +509,18 @@
             "extract": {
               "column": "human_subject_name"
             }
+          },
+          "references": {
+            "field": {
+              "@id": "subjects/human_subject_name"
+            }
           }
         },
         {
           "@type": "cr:Field",
           "@id": "synglass_sessions/eyeglass_model_name",
           "name": "eyeglass_model_name",
-          "description": "Anonymized eyeglass model identifier (glasses_001 through glasses_867).",
+          "description": "Anonymized eyeglass model identifier; joins eyeglasses.",
           "dataType": "sc:Text",
           "source": {
             "fileObject": {
@@ -202,13 +529,18 @@
             "extract": {
               "column": "eyeglass_model_name"
             }
+          },
+          "references": {
+            "field": {
+              "@id": "eyeglasses/eyeglass_model_name"
+            }
           }
         },
         {
           "@type": "cr:Field",
           "@id": "synglass_sessions/capture_session_index",
           "name": "capture_session_index",
-          "description": "Capture index from 1 to 3.",
+          "description": "Capture index 1-3 (three random head poses per avatar-eyeglass pair).",
           "dataType": "sc:Integer",
           "source": {
             "fileObject": {
@@ -223,7 +555,7 @@
           "@type": "cr:Field",
           "@id": "synglass_sessions/file_stem",
           "name": "file_stem",
-          "description": "Common anonymized filename stem shared by all resources in the session.",
+          "description": "Common filename stem of the four resources.",
           "dataType": "sc:Text",
           "source": {
             "fileObject": {
@@ -238,7 +570,7 @@
           "@type": "cr:Field",
           "@id": "synglass_sessions/color",
           "name": "color",
-          "description": "Color image, PNG, 1024x1024.",
+          "description": "Relative path of the 1024x1024 colour render (PNG).",
           "dataType": "sc:Text",
           "source": {
             "fileObject": {
@@ -253,7 +585,7 @@
           "@type": "cr:Field",
           "@id": "synglass_sessions/mask_frame",
           "name": "mask_frame",
-          "description": "Eyeglasses frame segmentation mask with depth encoding, RGBA PNG, 1024x1024.",
+          "description": "Relative path of the frame mask (RGBA PNG; black = not frame; non-black channels encode normalized depth).",
           "dataType": "sc:Text",
           "source": {
             "fileObject": {
@@ -268,7 +600,7 @@
           "@type": "cr:Field",
           "@id": "synglass_sessions/mask_lens",
           "name": "mask_lens",
-          "description": "Eyeglasses lens segmentation mask with depth encoding, RGBA PNG, 1024x1024.",
+          "description": "Relative path of the lens mask (RGBA PNG; black = not lens; non-black channels encode normalized depth).",
           "dataType": "sc:Text",
           "source": {
             "fileObject": {
@@ -283,7 +615,7 @@
           "@type": "cr:Field",
           "@id": "synglass_sessions/metadata",
           "name": "metadata",
-          "description": "Per-session JSON metadata file. Auxiliary fields such as avatar distance and head orientation are intentionally left inside this JSON file and are not indexed as Croissant fields.",
+          "description": "Relative path of the per-render JSON metadata (head pose, head/eye positions, camera parameters, avatar distance).",
           "dataType": "sc:Text",
           "source": {
             "fileObject": {
@@ -296,8 +628,5 @@
         }
       ]
     }
-  ],
-  "datePublished": "2026-05-01",
-  "citeAs": "SynGlass: Synthetic Eyeglasses Segmentation Dataset. TVARITA Research, 2026.",
-  "conformsTo": "http://mlcommons.org/croissant/1.1"
+  ]
 }