perf: recalibrate vram_plan (leave GPU headroom) + refresh perf README

README.md

@@ -37,15 +37,15 @@ impact-seg-konfai segment body -i input.nii.gz -o output/
 
 ## ⚡ Performance & VRAM
 
-Benchmarked on an **NVIDIA RTX PRO 5000 (24 GB)**. The batch size is **auto-selected from your free GPU VRAM**.
+Benchmarked on a single **NVIDIA RTX PRO 5000 (24 GB)** with a real whole-body CT (295 × 259 × 219, 2 mm). The batch size is **auto-selected from your free GPU VRAM**.
 
-| Free VRAM | Batch (auto) | Peak VRAM |
-|:--|:--|:--|
-| 8 GB  | 160 | ~7 GB |
-| 16 GB | 320 | ~14 GB |
-| 24 GB | 512 | ~22 GB |
+| Free VRAM | Batch (auto) | Peak VRAM | Time / case |
+|:--|:--|:--|:--|
+| 8 GB  | 160 | ~7 GB  | — |
+| 16 GB | 320 | ~14 GB | — |
+| 24 GB | 512 | ~10 GB | **~7 s** |
 
-≈ **16 s / case** on the benchmark volume (scales with case size). Override with `--patch-size` / `--batch-size`.
+Single-model body segmentation keeps **system RAM ~1.6 GB**. The thin 2-D patches never fill the card, so inference stays compute-bound (~7 s, largely batch-independent). Override with `--patch-size` / `--batch-size`.
 
 ## 🔗 Links