Refine DRF preprocessing and body-prior pipeline

datasets/Scoliosis1K/README.md

@@ -99,14 +99,22 @@ The PAV pass is implemented from the paper:
 4. compute vertical, midline, and angular deviations for the 8 symmetric joint pairs
 5. apply IQR filtering per metric
 6. average over time
-7. min-max normalize across the dataset, or across `TRAIN_SET` when `--stats_partition` is provided
+7. min-max normalize across the full dataset (paper default), or across `TRAIN_SET` when `--stats_partition` is provided as an anti-leakage variant
 
 Run:
 
 ```bash
 uv run python datasets/pretreatment_scoliosis_drf.py \
   --pose_data_path=<path_to_pose_pkl> \
-  --output_path=<path_to_drf_pkl> \
+  --output_path=<path_to_drf_pkl>
+```
+
+To reproduce the paper defaults more closely, the script now uses
+`configs/drf/pretreatment_heatmap_drf.yaml` by default, which enables
+summed two-channel skeleton maps and a literal 128-pixel height normalization.
+If you explicitly want train-only PAV min-max statistics, add:
+
+```bash
   --stats_partition=./datasets/Scoliosis1K/Scoliosis1K_118.json
 ```