Normalize all filenames to case_{cid:03d}_pt_{pt:03d}_frame_{frame:04d}_{tissue}.png; update metadata, splits, and README

README.md

@@ -25,8 +25,8 @@ Cystoscopic still frames (1 754 PNG, ~236 MB) sourced from a Kaggle dataset, org
 
 | Path | Description |
 |---|---|
-| `data/images/` | 1 754 PNG frames. Filenames follow four structural patterns (see below); `case_id` is extracted from the leading case number. |
-| `data/metadata.parquet` | One row per image. Columns: `filename` (actual disk name), `case_id` (int), `imaging_type` (WLI/NBI), `tissue_type` (HGC/LGC/NST/NTL), `original_split` (the **leaky** CSV split — kept for reference only), `csv_filename` (original CSV name, differs for HLT-renamed files). |
+| `data/images/` | 1 754 PNG frames. All filenames normalized to `case_{cid:03d}_pt_{pt:03d}_frame_{frame:04d}_{tissue}.png`. |
+| `data/metadata.parquet` | One row per image. Columns: `filename` (normalized disk name), `case_id` (int), `imaging_type` (WLI/NBI), `tissue_type` (HGC/LGC/NST/NTL), `original_split` (the **leaky** CSV split — kept for reference only), `csv_filename` (same as `filename` after normalization). |
 | `splits/{train,val,test}.txt` | Case-level image-id lists (seed=42). 18 / 2 / 2 cases → 1541 / 44 / 169 images. No case leakage. |
 
 ## Tissue types
@@ -38,11 +38,35 @@ Cystoscopic still frames (1 754 PNG, ~236 MB) sourced from a Kaggle dataset, org
 | HGC | 469 |
 | NTL | 134 |
 
+## Filename normalization
+
+The original Kaggle dataset contained 8 different filename patterns. All have
+been normalized to a single consistent format:
+
+```
+case_{cid:03d}_pt_{pt:03d}_frame_{frame:04d}_{tissue}.png
+```
+
+| Pattern | Original format | Count | Example |
+|---------|----------------|-------|---------|
+| 1 | `case_NNN_pt_NNN_frame_NNNN` | 1266 | `case_002_pt_003_frame_0009.png` |
+| 2 | `case_NNN_pt_NNN_HLT__frame_NNNN` | 7 | `case_012_pt_001_HLT__frame_0025.png` |
+| 3 | `case_NNN_pt_NNN_HLT_frame_NNNN` | 30 | `case_025_pt_004_HLT_frame_0000.png` |
+| 4 | `cys_case_N_ptN_frame_NNNN` | 28 | `cys_case_1_pt1_frame_2311.png` |
+| 5 | `cys_case_N_ptN_NNNN` | 262 | `cys_case_5_pt1_0055.png` |
+| 6 | `cys_case_N_ptN_NNNN (copy)` | 4 | `cys_case_10_pt1_1644 (copy).png` |
+| 7 | `cys_case_N_NNNN` (no pt) | 41 | `cys_case_7_0384.png` |
+| 8 | `case_N_cys_ptN_NNNN` | 116 | `case_6_cys_pt1_0165.png` |
+
+Normalization details:
+- **HLT annotation stripped** — 37 files had `HLT` (hyperplasia) embedded in the filename; already classified as `NST` in metadata, so the annotation was removed.
+- **Pattern 7 (no pt)** — 41 files had no patient number; assigned `pt_000`.
+- **" (copy)" suffix** — 4 files had macOS Finder duplicate suffixes; stripped (no non-copy counterparts existed; images are unique).
+- **Tissue type in filename** — 9 collision pairs existed where the same case/pt/frame had both cancer and non-cancer images; including `{tissue}` disambiguates them.
+
 ## Notes
 
 - **Original splits leak patients.** The CSV's `sub_dataset` column places 19 of 21 train cases also in val/test. This dataset ships new **case-level** splits (seed=42) to prevent leakage; `original_split` is preserved only for auditability.
-- **NST → HLT rename.** 37 files whose CSV name contains `_NST_` were renamed on disk to `_HLT_` (single or double underscore). `metadata.filename` is the **disk** name; `csv_filename` is the original CSV name.
-- **Filename patterns.** Four structural shapes appear (first regex match wins): `case_N_pt_N_*`, `case_N_cys_ptN_*`, `cys_case_N_ptN_*`, `cys_case_N_*`.
 
 ## Loading