Initial v1.0 dataset card + sensor-sync HDF5s + Croissant metadata

LICENSE

@@ -0,0 +1,30 @@
+Creative Commons Attribution 4.0 International (CC BY 4.0)
+
+Copyright (c) 2026 Rad-R Authors.
+
+You are free to:
+  - Share — copy and redistribute the material in any medium or format
+  - Adapt — remix, transform, and build upon the material
+  for any purpose, even commercially.
+
+The licensor cannot revoke these freedoms as long as you follow the license terms.
+
+Under the following terms:
+  - Attribution — You must give appropriate credit, provide a link to the
+    license, and indicate if changes were made. You may do so in any
+    reasonable manner, but not in any way that suggests the licensor
+    endorses you or your use.
+
+No additional restrictions — You may not apply legal terms or technological
+measures that legally restrict others from doing anything the license permits.
+
+Notices:
+  - You do not have to comply with the license for elements of the material
+    in the public domain or where your use is permitted by an applicable
+    exception or limitation.
+  - No warranties are given. The license may not give you all of the
+    permissions necessary for your intended use. For example, other rights
+    such as publicity, privacy, or moral rights may limit how you use the
+    material.
+
+Full license text: https://creativecommons.org/licenses/by/4.0/legalcode

README.md

@@ -0,0 +1,139 @@
+---
+license: cc-by-4.0
+task_categories:
+  - audio-classification
+  - time-series-forecasting
+language:
+  - en
+tags:
+  - radar
+  - mmwave
+  - raw-adc
+  - fault-diagnosis
+  - automotive
+  - sensor-fusion
+  - benchmark
+size_categories:
+  - 1K<n<10K
+pretty_name: "Rad-R: Raw-ADC mmWave Automotive Radar Robustness"
+configs:
+  - config_name: default
+    data_files:
+      - split: synced_hdf5
+        path: synced_hdf5/*.h5
+      - split: training_cache
+        path: training_cache.h5
+---
+
+# Rad-R: A Real-World Raw-ADC Dataset and Benchmark for mmWave Automotive Radar Robustness
+
+**Anonymous submission to NeurIPS 2026 — Evaluations & Datasets Track.**
+
+Rad-R is the first publicly released radar dataset combining **raw ADC captures** from a 4-chip TI MMWCAS-RF-EVM cascade radar (12 Tx × 16 Rx = 192 virtual channels, 77 GHz) with **physically induced hardware fault annotations** at multiple severity levels, synchronized with companion sensors (BNO055 IMU, DHT22 board/ambient temperature, three GPS streams, dual RPi cameras).
+
+## Captured fault classes (v1.0)
+
+| Class | Severity | Specification | File |
+|---|---|---|---|
+| Healthy | S0 | Baseline (no induced fault) | `healthy.h5` |
+| Misalignment | S1 (mild) | 5° yaw via precision jig + inclinometer | `yaw0.h5` |
+| Misalignment | S2 (severe) | 10° yaw | `yaw2.h5` |
+| Vibration | S1 (mild) | 20 Hz drive (eccentric DC motor on radar mount) | `vib1.h5` |
+| Vibration | S2 (severe) | 40 Hz drive | `vib2.h5` |
+| Blockage | S1 (mild) | 30 % polycarbonate coverage on PP radome | `blockage1.h5` |
+| Blockage | S2 (severe) | 60 % polycarbonate coverage on PP radome | `blockage2.h5` |
+| Rx degradation | S1 (mild) | 6/16 Rx antennas copper-taped | `degrade0.h5` |
+| Rx degradation | S2 (severe) | 10/16 Rx antennas copper-taped | `degrade1.h5` |
+
+Severity is gated by **measured physical quantities** (BNO055 RMS, inclinometer angle, corner-reflector dB drop, single-tone calibration sweep) — not just the procedural setting.
+
+## Repository layout
+
+```
+synced_hdf5/                    # Sensor-sync metadata, one HDF5 per capture (~25 MB total)
+  healthy.h5                    #   Per-radar-frame timestamps + IMU + DHT22 + GPS + camera paths
+  yaw0.h5  yaw2.h5
+  vib1.h5  vib2.h5
+  blockage1.h5  blockage2.h5
+  degrade0.h5  degrade1.h5
+training_cache.h5               # Pre-processed RD maps + raw IQ subset (~2.4 GB)
+                                #   1800 frames (200 sampled per capture)
+                                #   rd_map: (1800, 224, 224) f32 dB
+                                #   iq:     (1800, 64, 256, 16) complex64 (Tx=0 slice)
+                                #   fault_label, severity_label, capture, frame_idx
+LICENSE                         # CC BY 4.0
+croissant.json                  # ML metadata (NeurIPS Croissant standard)
+radr.py                         # HuggingFace `datasets` loader
+```
+
+## How to load (training cache, recommended)
+
+```python
+import h5py
+import numpy as np
+
+with h5py.File("training_cache.h5", "r") as f:
+    rd      = np.asarray(f["rd_map"])          # (1800, 224, 224) f32
+    iq      = np.asarray(f["iq"])              # (1800, 64, 256, 16) complex64
+    fault   = np.asarray(f["fault_label"])     # (1800,) int (0=healthy, 1=vib, 2=misalign, 3=block, 4=rxdeg)
+    sev     = np.asarray(f["severity_label"])  # (1800,) int (0=healthy, 1=mild, 2=severe)
+    capture = np.asarray(f["capture"]).astype(str)
+```
+
+## How to load (synced metadata + raw .bin pipeline)
+
+The 9 `synced_hdf5/*.h5` files contain per-radar-frame **wall-clock timestamps** and **synchronized companion-sensor streams**, but NOT the raw IQ — that lives in TI mmWave Studio `*_data.bin` capture directories (~324 GB total, available via Zenodo deposit on request).
+
+If you have the raw `.bin` directories alongside the synced HDF5s, the matlab-to-python pipeline (in the code repo) reads the cascaded ADC, applies range/Doppler FFT, and produces the same RD maps as in `training_cache.h5`.
+
+## Hardware
+
+| Spec | Value |
+|---|---|
+| Platform | TI MMWCAS-RF-EVM (4× AWR1243) |
+| Center frequency | 77 GHz |
+| Bandwidth | ~2.5 GHz (256 ADC samples × 79 MHz/μs) |
+| Tx × Rx | 12 × 16 = **192 virtual channels** |
+| Range resolution | 0.06 m |
+| Max range | 15.2 m |
+| Frame rate | 10 Hz (5 min × 9 captures = 45 min total recording) |
+| Radome | 1/16″ polypropylene panel (always installed) |
+
+## Companion sensors (per-frame aligned)
+
+- **Bosch BNO055 IMU** — 3-axis accel + gyro + temperature, ~33 Hz
+- **DHT22** — board + ambient temperature, ~1 Hz
+- **u-blox NEO-M9N GPS** — three decoded streams (legacy, bag-decoded, trips-DB-fused)
+- **Raspberry Pi cameras** — Cam2 + Cam4 PNG paths + frame timestamps
+
+## Evaluation tasks
+
+- **In-distribution fault classification** (5-way: healthy + 4 fault classes)
+- **Severity classification** (3-way: 0=healthy, 1=mild, 2=severe)
+- (Future: cross-day / cross-device / cross-scene OOD splits — deferred to v1.1 capture session)
+
+## Baselines benchmarked (in the paper)
+
+ResNet-18, ViT-Small, Raw IQ 1D-CNN, MambaRD, CLIP-B/32 linear probe, OpenCLIP ViT-L/14 linear probe, RadrNet (hierarchical Mamba on raw IQ).
+
+## License
+
+CC BY 4.0. You may use, distribute, and adapt this dataset, including for commercial purposes, provided you give attribution to the authors.
+
+## Citation
+
+```bibtex
+@inproceedings{radr2026,
+  title   = {Rad-R: A Real-World Raw-ADC Dataset and Benchmark for mmWave Automotive Radar Robustness},
+  author  = {Anonymous},
+  booktitle = {NeurIPS Evaluations \& Datasets Track},
+  year    = {2026}
+}
+```
+
+## Limitations (v1.0)
+
+- **Single capture device, single day, overlapping scenes.** Full OOD splits along device / day / scene axes deferred to a follow-on capture campaign.
+- **Two severity levels per fault** (mild / severe). The taxonomy supports up to 4 levels (S0–S3) but only S1–S2 are captured here.
+- **Thermal stress and RF interference** are part of the dataset specification but not yet captured. Scheduled for v1.1.
+- **Raw .bin not redistributed via this HF repo** (300 GB exceeds quota). Available on request; will be deposited on Zenodo at camera-ready.

croissant.json

@@ -0,0 +1,165 @@
+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "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"
+    },
+    "dct": "http://purl.org/dc/terms/",
+    "examples": {
+      "@id": "cr:examples",
+      "@type": "@json"
+    },
+    "extract": "cr:extract",
+    "field": "cr:field",
+    "fileProperty": "cr:fileProperty",
+    "fileObject": "cr:fileObject",
+    "fileSet": "cr:fileSet",
+    "format": "cr:format",
+    "includes": "cr:includes",
+    "isLiveDataset": "cr:isLiveDataset",
+    "jsonPath": "cr:jsonPath",
+    "key": "cr:key",
+    "md5": "cr:md5",
+    "parentField": "cr:parentField",
+    "path": "cr:path",
+    "recordSet": "cr:recordSet",
+    "references": "cr:references",
+    "regex": "cr:regex",
+    "repeated": "cr:repeated",
+    "replace": "cr:replace",
+    "sc": "https://schema.org/",
+    "separator": "cr:separator",
+    "source": "cr:source",
+    "subField": "cr:subField",
+    "transform": "cr:transform"
+  },
+  "@type": "sc:Dataset",
+  "name": "rad-r",
+  "description": "Rad-R: A Real-World Raw-ADC Dataset and Benchmark for mmWave Automotive Radar Robustness. The first publicly released radar dataset combining raw ADC captures from a 4-chip TI MMWCAS-RF-EVM cascade radar (12 Tx x 16 Rx, 77 GHz) with physically induced hardware fault annotations across 4 fault types and 2 severity levels each, plus a healthy baseline. Synchronized with BNO055 IMU, DHT22 temperature, three GPS streams, and dual RPi cameras. v1.0 contains 9 capture runs (5 min each, 10 fps) totalling ~27 k radar frames.",
+  "conformsTo": "http://mlcommons.org/croissant/1.0",
+  "url": "https://huggingface.co/datasets/radr-anon-2026/radr",
+  "version": "1.0.0",
+  "datePublished": "2026-05-07",
+  "license": "https://creativecommons.org/licenses/by/4.0/",
+  "citeAs": "@inproceedings{radr2026, title={Rad-R: A Real-World Raw-ADC Dataset and Benchmark for mmWave Automotive Radar Robustness}, author={Anonymous}, booktitle={NeurIPS Evaluations and Datasets Track}, year={2026}}",
+  "keywords": [
+    "radar",
+    "mmwave",
+    "raw-adc",
+    "fault diagnosis",
+    "automotive radar",
+    "sensor fusion",
+    "benchmark"
+  ],
+  "creator": {
+    "@type": "Organization",
+    "name": "Anonymous (NeurIPS 2026 submission)"
+  },
+
+  "rai:dataCollection": "Captured indoor and outdoor with a vehicle-mounted TI MMWCAS-RF-EVM (4-chip AWR1243 cascade, 77 GHz, 12 Tx x 16 Rx). Each fault was physically induced via dedicated hardware fixtures (precision yaw-offset jig with digital inclinometer, eccentric-mass DC vibration motor, polycarbonate sheet on PP radome, copper tape on Rx antennas). Severity is gated by measured physical quantities (IMU RMS, inclinometer angle, corner-reflector dB drop, single-tone calibration sweep). Capture protocol: each run is a continuous 5-minute recording at ~10 fps, with the same fault held throughout to provide stable training signal.",
+  "rai:dataAnnotationProtocol": "Per-clip labels (fault_class, severity) are entered once at capture-time by the operator into the RPi sensor-collection script; no manual frame-by-frame annotation. The companion sensors (IMU RMS, board-minus-ambient temperature, GPS speed) provide automatic per-frame regression-quality severity ground truth aligned to each radar frame via timestamp index lookups.",
+  "rai:dataReleaseMaintenancePlan": "Initial release: anonymized v1.0 hosted on HuggingFace (gated, auto-approve) for the NeurIPS double-blind review window. At camera-ready, ownership will be transferred to a stable academic organization account and the dataset will be made public. The full ~324 GB raw .bin captures will be deposited on Zenodo at camera-ready. Updates and v1.1 (with thermal-stress and RF-interference faults) will be tagged as separate releases.",
+  "rai:personalSensitiveInformation": "The RPi camera streams (Cam2 / Cam4) capture environmental scenes and may contain pedestrians or license plates. The released dataset includes only the camera frame paths and timestamps — the camera PNG frames themselves are NOT in this HuggingFace release. The radar data and companion-sensor streams contain no personally-identifiable information.",
+  "rai:dataLimitations": "v1.0 is captured with a single radar device on a single day, with overlapping scenes across faults. Genuine out-of-distribution evaluation along device / day / scene axes is therefore not supported by this release; planned for v1.1. Severity coverage is two levels per fault (mild and severe) rather than the three levels (S1, S2, S3) specified in the taxonomy; S3-level faults will be added in v1.1. Two of the six fault types in the taxonomy (thermal stress, RF interference) are not captured in v1.0.",
+  "rai:dataBiases": "Captures were collected with one operator on one day; no inter-operator variability is sampled in v1.0. Scene diversity is limited to the test environment used for the May 2026 collection session.",
+  "rai:dataUseCases": "Intended for academic research on (1) hardware fault diagnosis on mmWave automotive radar, (2) raw-ADC representation learning, (3) state-space modeling on complex IQ signals, and (4) sensor-fusion robustness under controlled degradation. Not intended as a substitute for safety-critical certification testing of production radar systems.",
+  "rai:dataSocialImpact": "Improves the reproducibility of radar fault diagnosis research and enables model-based health monitoring of automotive radars, contributing to safer ADAS / autonomous-driving systems. By releasing raw ADC under open license, lowers the barrier for academic groups without specialized data-collection hardware.",
+  "rai:dataSyntheticIndicator": "All captures are real physical recordings from a TI MMWCAS-RF-EVM. No synthetic data, no augmentation applied prior to release.",
+
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "training-cache",
+      "name": "training_cache.h5",
+      "description": "Pre-processed range-Doppler maps and raw IQ subset, 1800 frames (200 sampled per capture). Recommended entry point for ML training.",
+      "contentUrl": "training_cache.h5",
+      "encodingFormat": "application/x-hdf5",
+      "sha256": "TBD"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "synced-hdf5-set",
+      "name": "synced_hdf5",
+      "description": "9 sensor-sync HDF5 files, one per capture run. Per-radar-frame timestamps + IMU + DHT22 + GPS streams + camera paths.",
+      "encodingFormat": "application/x-hdf5",
+      "includes": "synced_hdf5/*.h5"
+    }
+  ],
+
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "@id": "training-cache-records",
+      "name": "training_cache_records",
+      "description": "Each record is one radar frame from one capture, with the RD map, the raw IQ slab (Tx=0 slice), and the fault/severity labels.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "rd_map",
+          "name": "rd_map",
+          "description": "Range-Doppler magnitude map, 224x224, dB scale, normalized per frame.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": { "@id": "training-cache" },
+            "extract": { "fileProperty": "content" }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "iq",
+          "name": "iq",
+          "description": "Raw complex IQ slab, shape (64 loops, 256 samples, 16 Rx), TDM-MIMO Tx=0 slice.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": { "@id": "training-cache" },
+            "extract": { "fileProperty": "content" }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "fault_label",
+          "name": "fault_label",
+          "description": "Fault class. 0=healthy, 1=vibration, 2=misalignment, 3=blockage, 4=rx_degradation.",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": { "@id": "training-cache" },
+            "extract": { "fileProperty": "content" }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "severity_label",
+          "name": "severity_label",
+          "description": "Severity level. 0=healthy baseline, 1=mild, 2=severe.",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": { "@id": "training-cache" },
+            "extract": { "fileProperty": "content" }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "capture",
+          "name": "capture",
+          "description": "Source capture identifier (one of healthy / yaw0 / yaw2 / vib1 / vib2 / blockage1 / blockage2 / degrade0 / degrade1).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": { "@id": "training-cache" },
+            "extract": { "fileProperty": "content" }
+          }
+        }
+      ]
+    }
+  ]
+}

radr.py

@@ -0,0 +1,100 @@
+"""HuggingFace `datasets` loader for Rad-R v1.0.
+
+Usage:
+    from datasets import load_dataset
+    ds = load_dataset("radr-anon-2026/radr", trust_remote_code=True)
+    sample = ds["train"][0]
+    # sample.keys() == {"rd_map", "iq_real", "iq_imag", "fault_label",
+    #                   "severity_label", "capture", "frame_idx"}
+"""
+from __future__ import annotations
+
+import datasets
+import h5py
+import numpy as np
+
+
+_CITATION = """\
+@inproceedings{radr2026,
+  title   = {Rad-R: A Real-World Raw-ADC Dataset and Benchmark for mmWave Automotive Radar Robustness},
+  author  = {Anonymous},
+  booktitle = {NeurIPS Evaluations and Datasets Track},
+  year    = {2026}
+}
+"""
+
+_DESCRIPTION = """\
+The first publicly released radar dataset combining raw ADC captures from a
+TI MMWCAS-RF-EVM 4-chip cascade (12 Tx x 16 Rx = 192 virtual channels at 77
+GHz) with physically induced hardware fault annotations. v1.0 contains 9
+capture runs (5 min each at 10 fps): healthy baseline, plus 4 fault types
+(vibration, misalignment, blockage, Rx degradation) at 2 severity levels each.
+Synchronized with BNO055 IMU, DHT22 board/ambient temperature, three GPS
+streams, and dual Raspberry Pi cameras.
+"""
+
+_HOMEPAGE = "https://huggingface.co/datasets/radr-anon-2026/radr"
+_LICENSE = "CC BY 4.0"
+
+_FAULT_NAMES = ["healthy", "vibration", "misalignment", "blockage", "rx_degradation"]
+_SEVERITY_NAMES = ["s0_healthy", "s1_mild", "s2_severe"]
+
+
+class RadRConfig(datasets.BuilderConfig):
+    def __init__(self, **kwargs):
+        super().__init__(version=datasets.Version("1.0.0"), **kwargs)
+
+
+class RadR(datasets.GeneratorBasedBuilder):
+    VERSION = datasets.Version("1.0.0")
+    BUILDER_CONFIGS = [
+        RadRConfig(name="default", description="Pre-processed RD maps + raw IQ subset (1800 frames)."),
+    ]
+    DEFAULT_CONFIG_NAME = "default"
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=datasets.Features({
+                "rd_map":         datasets.Array2D(shape=(224, 224), dtype="float32"),
+                "iq_real":        datasets.Array3D(shape=(64, 256, 16), dtype="float32"),
+                "iq_imag":        datasets.Array3D(shape=(64, 256, 16), dtype="float32"),
+                "fault_label":    datasets.ClassLabel(names=_FAULT_NAMES),
+                "severity_label": datasets.ClassLabel(names=_SEVERITY_NAMES),
+                "capture":        datasets.Value("string"),
+                "frame_idx":      datasets.Value("int32"),
+            }),
+            supervised_keys=("rd_map", "fault_label"),
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        cache_url = "training_cache.h5"
+        local = dl_manager.download(cache_url)
+        return [
+            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"path": local}),
+        ]
+
+    def _generate_examples(self, path):
+        with h5py.File(path, "r") as f:
+            n = int(f.attrs.get("n_total", f["fault_label"].shape[0]))
+            rd        = f["rd_map"]
+            iq        = f["iq"]
+            fault     = f["fault_label"]
+            severity  = f["severity_label"]
+            capture   = f["capture"]
+            frame_idx = f["frame_idx"]
+
+            for i in range(n):
+                iq_complex = iq[i]
+                yield i, {
+                    "rd_map":         rd[i],
+                    "iq_real":        iq_complex.real.astype(np.float32),
+                    "iq_imag":        iq_complex.imag.astype(np.float32),
+                    "fault_label":    int(fault[i]),
+                    "severity_label": int(severity[i]),
+                    "capture":        capture[i].decode() if isinstance(capture[i], bytes) else str(capture[i]),
+                    "frame_idx":      int(frame_idx[i]),
+                }

synced_hdf5/blockage1.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ad882cc5d5b2755e6ed13ede4e2dc18bb3a1e4a2b6270ca247af6db6fa88326
+size 2412822

synced_hdf5/blockage2.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8cb87168b64a9672f29ca158f9fc155b3d7ed3115500c3e24aa840e963c8a7ea
+size 3132962

synced_hdf5/degrade0.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4db4bbd4f5c7796093783a777428eb34504ccd2bc8dae51453f64ddd50af525
+size 3521664

synced_hdf5/degrade1.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e5f2022627979edc4db3856e0f83938d4e091196231d3122ef10191412c845e
+size 2930247

synced_hdf5/healthy.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4fa7d8863c77a88ac95b0a423abf10ef0ede5ae232bec88143254d8b0b8b5119
+size 3639270

synced_hdf5/vib1.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f6add28c2f7f8b54275f3fe13c29a24f951b3a558fe8bbeae1472a586951a07
+size 2345411

synced_hdf5/vib2.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bfe721ec6d1c507c08c1d51d2b3ba491938e9337d3531342132f394de6bd8dd0
+size 3072508

synced_hdf5/yaw0.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f08a2ffd1542a1c01bef080729ebe464e9bb7c3a708c32d67bba9801a5bf842
+size 2713149

synced_hdf5/yaw2.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c425c423f2f7e629e7d4b81a04cde8e79836cf135199a24b20297e4f61180a2
+size 2565401