Update README + croissant (simplified, 400-case framing, Croissant validation PASS)

README.md

@@ -1,108 +1,53 @@
-# Apple-Pi: Physics Infographics Evaluation Subset
+---
+license: cc-by-sa-4.0
+---
 
-A 100-case evaluation subset (50 simulation + 50 realworld) for physics-infographic reasoning evaluation in image and video generation models.
+# Apple-Pi
 
-## Overview
+A 100-case evaluation subset (50 simulation + 50 realworld) drawn from a larger **400-case physics benchmark** for evaluating physics reasoning in image and video generation models.
 
-This dataset contains ground-truth recordings and annotations for physics scenes spanning 9 simulated tasks and 7 realworld categories. Each case provides full per-frame rendering, instance segmentation, velocity tracking, depth, mask, and human-readable physical-parameter annotations, suitable for evaluating multi-step physics understanding in generative models.
-
-## Folder Structure
+## Folder structure
 
 ```
 apple-pi/
 ├── README.md
-├── croissant.json                  # Croissant + RAI metadata
-├── selected_cases.json             # Manifest with case IDs and sampling strategies
-├── sim_subset/                     # 50 simulation cases
-│   ├── at_rest/0/, at_rest/2/, ...
-│   ├── circular/0/, circular/newbatch_5/, ...
-│   └── ... (9 physics tasks)
-└── realworld_subset/               # 50 realworld cases
-    ├── Composition/0/
-    ├── Gravity/Freefall/1/
-    ├── ConservationofMomentum/InelasticCollision/1/
-    └── NewtonsFirstLaw/AtRest/1/
+├── croissant.json
+├── selected_cases.json
+├── sim_subset/        # 50 simulation cases
+└── realworld_subset/  # 50 realworld cases
 ```
 
-## Per-Case Contents (Simulation)
-
-Each `sim_subset/<task>/<case_id>/`:
-
-| File / Folder | Description |
-|---|---|
-| `annotation.txt` | Human-readable physical parameters (mass, velocity, friction, etc.) |
-| `physics_duration.txt` | Total physics-valid duration in seconds |
-| `formula_info.json` | 4 candidate formulas + correct answer + variable substitution hints |
-| `prompt.txt` | Original task prompt |
-| `initial_state/` | First-frame data: `rgb_0000.png` (with annotations), `rgb_0000_white_bg.png`, `rgb_0000_white_bg_obj.png` (object only on white), instance segmentation, mapping, depth, density, mask, velocity, camera parameters |
-| `instantaneous_velocity/` | Per-object velocity at a target time: `velocity.json`, `velocity_annotated.png`, `mask.npy`, `mapping.json` |
-| `instance_segmentation/maps.npz` | Full-sequence instance segmentation maps |
-| `mask/maps.npz` | Full-sequence foreground masks |
-| `velocity/maps.npz` | Full-sequence per-pixel velocity maps |
-| `depth/maps.npz` | Full-sequence depth maps |
-| `density/`, `camera_parameters/` | Per-frame density buffers and camera parameters |
-| `rgb/0000.png ~ 00NN.png` | Rendered video frames at 24 fps (NN = round(physics_duration × 24)) |
-
-## Per-Case Contents (Realworld)
-
-Each `realworld_subset/<category>/[<subcategory>/]<case_id>/`:
-
-| File / Folder | Description |
-|---|---|
-| `annotation.txt` | Physical parameters of the recorded scene |
-| `annotation_config.json` | Parameter source / unit metadata |
-| `formula_info.json` | Same 4-choice formula schema as simulation |
-| `physics_duration.txt` | Validated motion duration |
-| `initial_state/` | First-frame extraction: `rgb_0000.png`, `rgb_0000_white_bg.png`, `rgb_0000_white_bg_obj.png`, instance segmentation, mapping, mask |
-| `instance_segmentation/maps.npz` | Per-frame instance segmentation |
-| `mask/maps.npz` | Per-frame foreground mask |
-| `instantaneous_velocity/` | Velocity reference at target time: `velocity.json`, `velocity_annotated.png`, `mask.npy` |
-| `rgb/video.mp4` | Original video clip |
+## Per-case contents
 
-## Tasks Covered
+**Simulation** (`sim_subset/<task>/<case_id>/`):
+- `annotation.txt`, `physics_duration.txt`, `formula_info.json`
+- `initial_state/` (rgb_0000 + white-bg variants + instance segmentation + camera params)
+- `instantaneous_velocity/` (target-time velocity reference)
+- `rgb/0000.png ~ 00NN.png` (rendered video frames @ 24 fps)
+- `instance_segmentation/`, `mask/`, `velocity/`, `depth/`, `density/`, `camera_parameters/`
 
-### Simulation (9 physics tasks, 50 cases)
-- `at_rest`, `circular`, `freefall`, `inclined_plane`
-- `inelastic_collision`, `perfectly_elastic_collision`, `perfectly_inelastic_collision`
-- `projectile`, `uniform_linear`
+**Realworld** (`realworld_subset/<category>/[<subcategory>/]<case_id>/`):
+- `annotation.txt`, `formula_info.json`, `physics_duration.txt`, `annotation_config.json`
+- `initial_state/` (first-frame extraction + variants)
+- `instance_segmentation/`, `mask/`, `instantaneous_velocity/`
+- `rgb/video.mp4`
 
-### Realworld (7 categories, 50 cases)
-- `NewtonsFirstLaw/AtRest`
-- `Gravity/Freefall`, `Gravity/Projectile`
-- `ConservationofMomentum/{InelasticCollision, PerfectlyElasticCollision, PerfectlyInelasticCollision}`
-- `Composition` (multi-task)
+## Tasks
 
-## Sampling
+**Simulation (9 tasks)**: at_rest, circular, freefall, inclined_plane, inelastic_collision, perfectly_elastic_collision, perfectly_inelastic_collision, projectile, uniform_linear.
 
-- **Simulation**: 50 cases selected from a larger 100-case pool with stratified sampling (seed=42, OLD-first); excludes any cases with rendering artifacts.
-- **Realworld**: 50 cases sampled from a 157-case pool with stratified sampling proportional to per-subcategory counts (seed=42).
+**Realworld (7 categories)**: NewtonsFirstLaw/AtRest, Gravity/{Freefall, Projectile}, ConservationofMomentum/{InelasticCollision, PerfectlyElasticCollision, PerfectlyInelasticCollision}, Composition.
 
-See `selected_cases.json` for full case-ID lists and sampling metadata.
+## Intended use
 
-## Intended Use
-
-- Evaluating image-generation and video-generation models on physics reasoning tasks (perception, comprehension, generation).
-- Comparing model outputs against ground-truth simulation rollouts and real-video frames.
-- Studying instruction-following and prompt-sensitivity in multimodal models.
-
-## Dataset Composition
-
-| Domain | Cases | Avg files/case | Approx total files |
-|---|---:|---:|---:|
-| Simulation | 50 | ~750 | ~37,000 |
-| Realworld | 50 | ~16 | ~800 |
-| **Total** | **100** | | ~37,800 |
+Zero-shot evaluation of physics reasoning in image- and video-generation models (perception, comprehension, generation tracks). Not intended for training.
 
 ## License
 
-Released under **CC BY-SA 4.0** (Creative Commons Attribution–ShareAlike 4.0 International).
-
-## Anonymity
-
-This dataset is released for double-blind submission. All identifying information has been removed.
+CC BY-SA 4.0
 
 ## Citation
 
 ```
-[Anonymous] Apple-Pi: Physics Infographics Evaluation Subset. 2026.
+[Anonymous] Apple-Pi: Physics Infographics Benchmark. 2026.
 ```

croissant.json

@@ -7,19 +7,10 @@
     "conformsTo": "dct:conformsTo",
     "cr": "http://mlcommons.org/croissant/",
     "rai": "http://mlcommons.org/croissant/RAI/",
-    "data": {
-      "@id": "cr:data",
-      "@type": "@json"
-    },
-    "dataType": {
-      "@id": "cr:dataType",
-      "@type": "@vocab"
-    },
+    "data": {"@id": "cr:data", "@type": "@json"},
+    "dataType": {"@id": "cr:dataType", "@type": "@vocab"},
     "dct": "http://purl.org/dc/terms/",
-    "examples": {
-      "@id": "cr:examples",
-      "@type": "@json"
-    },
+    "examples": {"@id": "cr:examples", "@type": "@json"},
     "extract": "cr:extract",
     "field": "cr:field",
     "fileProperty": "cr:fileProperty",
@@ -38,6 +29,8 @@
     "regex": "cr:regex",
     "repeated": "cr:repeated",
     "replace": "cr:replace",
+    "samplingRate": "sc:samplingRate",
+    "equivalentProperty": "sc:equivalentProperty",
     "sc": "https://schema.org/",
     "separator": "cr:separator",
     "source": "cr:source",
@@ -46,64 +39,31 @@
   },
   "@type": "sc:Dataset",
   "name": "apple-pi",
-  "description": "A 100-case evaluation subset (50 simulation + 50 realworld) for physics-infographic reasoning evaluation in image and video generation models. Each case provides ground-truth rendering, instance segmentation, velocity tracking, depth, mask, and human-readable physics annotations across 9 simulated tasks and 7 realworld categories.",
+  "description": "Apple-Pi is a 100-case evaluation subset of a larger 400-case physics benchmark for evaluating physics reasoning in image and video generation models. Each case provides a video clip (or rendered frames) with ground-truth annotations of physical parameters, instance segmentation, and per-frame velocity. Released for double-blind submission.",
   "conformsTo": "http://mlcommons.org/croissant/1.0",
   "license": "https://creativecommons.org/licenses/by-sa/4.0/",
   "url": "https://huggingface.co/datasets/AnonymousSubmitt/apple-pi",
   "version": "1.0.0",
   "datePublished": "2026-05-06",
-  "citeAs": "@misc{anonymous2026applepi, title={Apple-Pi: Physics Infographics Evaluation Subset}, author={Anonymous}, year={2026}}",
-  "creator": {
-    "@type": "Organization",
-    "name": "Anonymous"
-  },
-  "publisher": {
-    "@type": "Organization",
-    "name": "Anonymous"
-  },
-  "keywords": [
-    "physics reasoning",
-    "video generation evaluation",
-    "image generation evaluation",
-    "physics infographics",
-    "instance segmentation",
-    "ground truth",
-    "synthetic + realworld"
-  ],
-  "rai:dataCollection": "Synthetic 3D physics simulation rendered with a real-time physics engine and rasterized rendering pipeline. Realworld videos recorded under controlled conditions with annotated physical parameters.",
-  "rai:dataCollectionType": ["Synthetic data", "Recorded video"],
-  "rai:dataCollectionTimeframe": "2025-2026",
-  "rai:annotationsSourceProcess": "Simulation cases: programmatically annotated from physics-engine state (deterministic per-frame instance segmentation, velocity, depth, mask). Realworld cases: human-labeled physical parameters with computed segmentation and target-time velocity.",
-  "rai:personalSensitiveInformation": "None. Synthetic cases contain no people. Realworld videos do not show identifiable individuals.",
-  "rai:bias": "Synthetic data introduces no demographic bias. Realworld videos contain only inanimate objects. The case distribution is intentionally weighted toward common physics phenomena (projectile, freefall) which may bias evaluation toward those tasks.",
-  "rai:knownLimitations": "Synthetic data may not capture all real-world physical complexity (air drag, non-rigid deformation, lighting variation). Realworld videos are limited to short clips of constrained motion. The 100-case subset is intentionally small for fast evaluation, not for training.",
-  "rai:intendedUse": "Evaluating physics reasoning capabilities of image- and video-generation models, including perception, comprehension, and generation tracks. Comparing model outputs against ground-truth physics rollouts and real-video frames. Studying instruction-following and prompt-sensitivity.",
-  "rai:useCases": [
-    "Benchmarking video-generation models (e.g. Veo, Sora) on physics-grounded prompts",
-    "Benchmarking unified image-generation models on physics-grounded prompts",
-    "Studying prompt-complexity ablations and instruction-following"
-  ],
-  "rai:potentialApplications": "Research evaluation of multimodal generation models in physics reasoning.",
-  "rai:misuse": "Low risk. Dataset contains no human subjects or sensitive content. Misuse limited to inappropriate benchmarking claims.",
-  "rai:safetyMeasures": "Dataset is small and well-scoped; intended only for evaluation, not training. Cases are physics-only with no human or social context.",
-  "rai:maintenanceStatus": "Active",
-  "rai:dataReleases": "v1.0 (2026-05): Initial release with 50 sim + 50 realworld cases.",
+  "citeAs": "@misc{anonymous2026applepi, title={Apple-Pi: Physics Infographics Benchmark}, author={Anonymous}, year={2026}}",
+  "creator": {"@type": "Organization", "name": "Anonymous"},
+  "keywords": ["physics reasoning", "video generation", "image generation", "benchmark", "evaluation"],
+
+  "rai:dataLimitations": "This subset is intentionally small (100 cases) and used only for evaluation, not training. Synthetic cases use a rigid-body physics engine and do not capture deformation or fluid dynamics. Realworld cases are short controlled clips with limited environmental diversity. Not recommended for safety-critical or clinical applications.",
+  "rai:dataBiases": "Task distribution is weighted toward common phenomena (projectile, freefall) and may under-represent rarer tasks (composition, multi-body collision). Synthetic cases use a fixed object/material library; realworld cases are recorded under controlled lighting. No demographic bias because no humans are present.",
+  "rai:personalSensitiveInformation": "None. Synthetic cases contain no people; realworld videos do not show identifiable individuals.",
+  "rai:dataUseCases": "Evaluating physics reasoning in image- and video-generation models across perception, comprehension, and generation tracks. Validated for zero-shot benchmark evaluation. Not validated for training, clinical decision-making, or robotic control.",
+  "rai:dataSocialImpact": "Positive: provides a controlled benchmark to measure physics understanding gaps in current generation models. Risk: scores may be over-interpreted as general physics reasoning ability. Mitigation: released under CC BY-SA 4.0 with explicit research-evaluation framing; contains no humans or sensitive content.",
+  "rai:dataCollection": "Synthetic cases generated with a rigid-body 3D physics engine; ground truth computed directly from physics-engine state. Realworld cases recorded as short video clips under controlled laboratory conditions; physical parameters human-labeled, segmentation and velocity computed via standard tracking pipelines.",
+  "rai:annotationsPerItem": "Each case includes physical parameters (annotation.txt), governing-equation choices with correct answer (formula_info.json), per-object instance segmentation, target-time velocity reference, and per-frame mask/depth/density (synthetic only).",
+  "rai:dataReleases": "v1.0 (2026-05): initial release of 100-case evaluation subset.",
+
   "distribution": [
-    {
-      "@type": "cr:FileObject",
-      "@id": "repo",
-      "name": "repository",
-      "description": "Hugging Face dataset repository.",
-      "contentUrl": "https://huggingface.co/datasets/AnonymousSubmitt/apple-pi",
-      "encodingFormat": "git+https",
-      "sha256": "main"
-    },
     {
       "@type": "cr:FileSet",
       "@id": "sim_subset",
       "name": "sim_subset",
-      "description": "50 simulation cases. Each case is a folder containing rendered video frames, instance segmentation, velocity, depth, mask, and physics annotations.",
-      "containedIn": {"@id": "repo"},
+      "description": "50 simulation cases. Each case is a folder containing rendered video frames, instance segmentation, depth, mask, velocity, and physics annotations.",
       "encodingFormat": "application/octet-stream",
       "includes": "sim_subset/**"
     },
@@ -111,56 +71,9 @@
       "@type": "cr:FileSet",
       "@id": "realworld_subset",
       "name": "realworld_subset",
-      "description": "50 realworld cases. Each case is a folder containing a video clip, instance segmentation, velocity, mask, and physics annotations.",
-      "containedIn": {"@id": "repo"},
+      "description": "50 realworld cases. Each case is a folder containing a video clip, instance segmentation, mask, velocity reference, and physics annotations.",
       "encodingFormat": "application/octet-stream",
       "includes": "realworld_subset/**"
     }
-  ],
-  "recordSet": [
-    {
-      "@type": "cr:RecordSet",
-      "@id": "sim_cases",
-      "name": "Simulation Cases",
-      "description": "Each record is one simulation case folder with a fixed-format internal layout.",
-      "field": [
-        {
-          "@type": "cr:Field",
-          "@id": "sim_cases/case_path",
-          "name": "case_path",
-          "description": "Relative folder path inside sim_subset/ (e.g. circular/0).",
-          "dataType": "sc:Text"
-        },
-        {
-          "@type": "cr:Field",
-          "@id": "sim_cases/task",
-          "name": "task",
-          "description": "Physics task name (top-level directory under sim_subset/).",
-          "dataType": "sc:Text"
-        }
-      ]
-    },
-    {
-      "@type": "cr:RecordSet",
-      "@id": "realworld_cases",
-      "name": "Realworld Cases",
-      "description": "Each record is one realworld case folder with a fixed-format internal layout.",
-      "field": [
-        {
-          "@type": "cr:Field",
-          "@id": "realworld_cases/case_path",
-          "name": "case_path",
-          "description": "Relative folder path inside realworld_subset/ (e.g. Gravity/Freefall/1).",
-          "dataType": "sc:Text"
-        },
-        {
-          "@type": "cr:Field",
-          "@id": "realworld_cases/category",
-          "name": "category",
-          "description": "Physics category (top-level directory).",
-          "dataType": "sc:Text"
-        }
-      ]
-    }
   ]
 }