Update dataset card after external splits upload: README.md

README.md

@@ -30,18 +30,27 @@ configs:
 
 ## Dataset Summary
 
-TODO: Describe MMPD-Bench, the scientific motivation, and the intended benchmark
-use. Mention that the dataset contains paired Mueller matrix observations and
-Lu-Chipman reference decomposition outputs.
+MMPD-Bench is a polarimetric imaging benchmark for learning mappings from
+Mueller matrix observations to polarimetric decomposition modalities. Each
+sample contains a channel-first Mueller matrix tensor and a channel-first target
+tensor with six Lu-Chipman reference modalities.
 
-Current pilot upload: external test splits only. Use config `external_waveplate`
-for the 633 nm waveplate split and config `external_spectral` for the
-610/650/690 nm spectral splits.
+Current Hugging Face release status:
+
+- Uploaded: external waveplate test data at 633 nm.
+- Uploaded: external spectral test data at 610, 650, and 690 nm.
+- Not yet uploaded: healthy bone cell `train`, `validation`, and `test` splits.
+
+Because the waveplate tensors are 200 x 200 while the spectral tensors are
+256 x 256, the external data is published as two separate configs:
+
+- `external_waveplate`
+- `external_spectral`
 
 ## Task Definition
 
-TODO: Define the task as modality fission from a Mueller matrix patch to six
-polarimetric target modalities.
+The task is modality fission from a Mueller matrix tensor to six polarimetric
+target modalities. It is not a segmentation or classification dataset.
 
 - Input: Mueller matrix tensor, shape `[16, H, W]`, channel-first.
 - Output: target modality tensor, shape `[6, H, W]`, channel-first.
@@ -49,9 +58,14 @@ polarimetric target modalities.
 
 ## Data Sources
 
-TODO: Describe healthy bone cell samples, waveplate samples, and
-multi-wavelength measurements. Include acquisition context and de-identification
-statement if applicable.
+This release contains external test data from `polarization_v3`:
+
+- Waveplate data: `hwp633` and `qwp633`, measured at 633 nm.
+- Multi-wavelength spectral data: selected wavelengths from `mwl_530_690`,
+  currently 610, 650, and 690 nm.
+
+The healthy bone cell data from `polarization_v2` is planned for a later upload
+as `train`, `validation`, and `test`.
 
 ## File Structure
 
@@ -59,41 +73,62 @@ statement if applicable.
 MMPD-Bench/
 ├── README.md
 ├── data/
-├── metadata/
-├── preview/
-└── raw/
+│   ├── external_waveplate-00000-of-00001.parquet
+│   ├── external_spectral_610-00000-of-00001.parquet
+│   ├── external_spectral_650-00000-of-00001.parquet
+│   └── external_spectral_690-00000-of-00001.parquet
+└── metadata/
+    ├── acquisition_info.json
+    ├── channel_order.json
+    ├── parameter_ranges.json
+    ├── schema.json
+    └── split_summary.json
 ```
 
-TODO: Replace this with final shard counts and file sizes after conversion.
-
 ## Tensor Schema
 
-Each row represents one paired patch/sample:
+Common columns:
 
 ```python
 {
     "sample_id": str,
     "source_id": str,
     "split": str,
-    "subset": str,         # healthy_bone_cell, waveplate, spectral
-    "specimen_type": str,  # healthy_bone_cell, waveplate, spectral, unknown
+    "subset": str,          # waveplate or spectral in the current release
+    "specimen_type": str,   # waveplate or spectral in the current release
     "wavelength_nm": int,
-    "patch_id": str,  # external_spectral_* rows
-    "target_encoding": str,
+    "source_path": str,
     "mueller_shape": list[int],
     "target_shape": list[int],
-    "mueller": array,  # float32, channel-first
-    "target": array,   # float32, channel-first
+    "mueller": array,       # float32, channel-first
+    "target": array,        # float32, channel-first
+}
+```
+
+Waveplate-specific columns:
+
+```python
+{
+    "plate_type": str,      # hwp or qwp
+    "angle_label": str,     # e.g. 0deg, n22, p45
+    "angle_deg": float,
 }
 ```
 
-TODO: Confirm whether all future arrays are stored as physical values,
-normalized values, or both. Current pilot encoding:
+Spectral-specific columns:
 
-- `external_waveplate`: tensors have shapes `[16, 200, 200]` and `[6, 200, 200]`.
-- `external_spectral_*`: tensors have shapes `[16, 256, 256]` and
-  `[6, 256, 256]`; targets are grayscale PNG values normalized to float32
-  `[0, 1]`.
+```python
+{
+    "patch_id": str,
+    "target_encoding": str, # png_uint8_normalized_to_float32_0_1
+}
+```
+
+Current tensor shapes:
+
+- `external_waveplate`: `mueller = [16, 200, 200]`, `target = [6, 200, 200]`.
+- `external_spectral_*`: `mueller = [16, 256, 256]`,
+  `target = [6, 256, 256]`.
 
 ## Channel Conventions
 
@@ -112,12 +147,12 @@ Target channel order:
 D, Delta, eta, theta, psi, R
 ```
 
-TODO: Document the mapping from local filenames such as `Ita`, `ita`, or `Eta`
-to public channel name `eta`.
+Local source files may use names such as `Ita`, `ita`, or `Eta`; the public
+channel name is normalized to `eta`.
 
 ## Physical Parameter Definitions
 
-TODO: Define every target parameter and confirm ranges.
+The target tensor follows this channel order and nominal parameter range:
 
 ```text
 D, Delta: [0, 1]
@@ -125,81 +160,113 @@ eta, R: [0, pi)
 theta, psi: [-pi/2, pi/2)
 ```
 
-## Reference Label Generation
+Important encoding note:
+
+- Waveplate target arrays are stored from the source `.npy` files as float32.
+- Spectral target arrays were converted from grayscale PNG files to float32
+  values normalized to `[0, 1]`; see `target_encoding`.
 
-The target modalities are generated using Lu-Chipman decomposition from the
-measured Mueller matrices. They should be interpreted as reference solver
-outputs for benchmarking surrogate models and physics consistency, rather than
-direct human annotations or absolute ground truth for biological tissue.
+## Reference Label Generation
 
-TODO: Add decomposition implementation details, preprocessing, and any numerical
-stability handling.
+The target modalities are generated using Lu-Chipman decomposition from measured
+Mueller matrices. They should be interpreted as physics-solver reference labels
+for benchmarking surrogate models and physics consistency, not as direct human
+annotations or absolute biological ground truth.
 
 ## Splits
 
-TODO: Fill exact counts after conversion.
+| Split | Config | Subset | Samples | Shape | Notes |
+|---|---|---:|---:|---|---|
+| external_waveplate | external_waveplate | waveplate | 18 | `[16, 200, 200] -> [6, 200, 200]` | 633 nm HWP/QWP |
+| external_spectral_610 | external_spectral | spectral | 165 | `[16, 256, 256] -> [6, 256, 256]` | 610 nm |
+| external_spectral_650 | external_spectral | spectral | 165 | `[16, 256, 256] -> [6, 256, 256]` | 650 nm |
+| external_spectral_690 | external_spectral | spectral | 165 | `[16, 256, 256] -> [6, 256, 256]` | 690 nm |
+| train | not uploaded yet | healthy_bone_cell | TBD | TBD | planned |
+| validation | not uploaded yet | healthy_bone_cell | TBD | TBD | planned |
+| test | not uploaded yet | healthy_bone_cell | TBD | TBD | planned |
 
-| Split | Subset | Samples | Notes |
-|---|---:|---:|---|
-| train | healthy_bone_cell | TODO | Primary training split |
-| validation | healthy_bone_cell | TODO | Model selection split |
-| test | healthy_bone_cell | TODO | Clean healthy bone cell test |
-| external_waveplate | waveplate | 18 | External generalisation at 633 nm |
-| external_spectral_610 | spectral | 165 | External wavelength test |
-| external_spectral_650 | spectral | 165 | External wavelength test |
-| external_spectral_690 | spectral | 165 | External wavelength test |
+## Benchmark Protocols
 
-TODO: State whether healthy bone cell splits are specimen-level or patch-level.
-If patch-level, describe leakage risk as a limitation.
+Current release:
 
-## External Test Sets
+1. External waveplate evaluation: use config `external_waveplate`, split
+   `external_waveplate`.
+2. External spectral evaluation: use config `external_spectral`, then evaluate
+   `external_spectral_610`, `external_spectral_650`, and
+   `external_spectral_690`.
 
-TODO: Describe waveplate and spectral external test sets, including wavelength,
-sample source, and intended usage.
+Planned full benchmark:
 
-## Benchmark Protocols
+1. Train on healthy bone cell `train`.
+2. Tune on healthy bone cell `validation`.
+3. Report clean performance on healthy bone cell `test`.
+4. Report external generalisation on waveplate and spectral splits.
 
-1. Clean healthy bone cell test: train on `train`, tune on `validation`, report on
-   `test`.
-2. Noisy healthy bone cell robustness: evaluate the test set with
-   `sigma_noise = 0.1 * sigma_pixel`.
-3. External generalisation: train only on healthy bone cell training data and
-   report on `external_waveplate` and each `external_spectral_*` split.
+## Loading Instructions
 
-## Metrics
+Install the Hugging Face datasets package:
 
-TODO: List recommended metrics for each target modality and any aggregate score.
+```bash
+pip install datasets
+```
 
-## Loading Instructions
+Load one external spectral split:
 
 ```python
 from datasets import load_dataset
-
-ds = load_dataset("YOUR_ORG/MMPD-Bench")
-print(ds)
-print(ds["train"][0].keys())
+import numpy as np
+
+ds = load_dataset(
+    "parquet",
+    data_files={
+        "external_spectral_610": (
+            "hf://datasets/HY2333/MMPD_Bench/"
+            "data/external_spectral_610-*.parquet"
+        )
+    },
+    split="external_spectral_610",
+)
+
+row = ds[0]
+mueller = np.array(row["mueller"], dtype=np.float32)
+target = np.array(row["target"], dtype=np.float32)
+
+print(row["sample_id"])
+print(mueller.shape)
+print(target.shape)
 ```
 
+Load via dataset config:
+
 ```python
-import torch
+from datasets import load_dataset
 
-row = ds["train"][0]
-mueller = torch.tensor(row["mueller"], dtype=torch.float32)
-target = torch.tensor(row["target"], dtype=torch.float32)
+spectral = load_dataset("HY2333/MMPD_Bench", "external_spectral")
+waveplate = load_dataset("HY2333/MMPD_Bench", "external_waveplate")
 ```
 
+Note: in some environments, streaming reads of large nested Parquet tensors can
+trigger a PyArrow shutdown issue after successful iteration. For a stable smoke
+test, use non-streaming loading on a single split as shown above.
+
 ## Ethics and Limitations
 
-TODO: Add de-identification statement, biological ground-truth limitation, and
-recommended use boundaries.
+The current release contains external physical/spectral evaluation data, not the
+healthy bone cell train/validation/test splits. Biological-data ethics and
+de-identification details should be completed before publishing the healthy bone
+cell splits.
+
+The targets are Lu-Chipman reference outputs. Evaluation should be interpreted
+as agreement with a physics-solver reference and related physics consistency,
+not as proof of absolute biological ground truth.
 
 ## License
 
-TODO: Add final dataset license and any access restrictions.
+This dataset is released under CC BY-NC 4.0.
 
 ## Citation
 
-TODO: Add paper citation and BibTeX.
+TODO: Add the MMPD-Bench paper citation and BibTeX entry.
 
 ## Contact