update docs for prev pointing changes

CHANGELOG.md

@@ -3,6 +3,14 @@
 All notable changes to **`nvidia/PhysicalAI-VANTAGE-Bench`** on Hugging Face.
 
 ## 2026-05-27
+- **`data/pointing/VANTAGE_2DPointing.tsv` updated.** MCQ option
+  coordinates (columns `A`–`D`) are now expressed in the normalized
+  `0–1000` coordinate system (each cell is an `x,y` pair, with both
+  components in `[0, 1000]` relative to the image dimensions). The
+  public TSV ships only the question side: columns are
+  `index, question_id, image_path, question, A, B, C, D`; the
+  ground-truth fields (`answer`, `target_point`) are held server-side
+  and are not included in the released file.
 - **`data/event_verification/` flattened** to match the `vqa/` and
   `temporal_localization/` layout. The `filtered/.../{metropolis,tailgating,warehouse_near_miss}`
   subtree was removed; all 163 videos now live directly under

README.md

@@ -22,7 +22,7 @@ configs:
 - config_name: pointing
   data_files:
   - split: test
-    path: data/pointing/Vantage2DPointing.tsv
+    path: data/pointing/VANTAGE_2DPointing.tsv
 - config_name: tracking
   data_files:
   - split: test

data/README.md

@@ -18,7 +18,7 @@ data/
 │   ├── *.mp4 (under videos/)
 │   └── data_jsons/annotations/*.json
 ├── pointing/                     # 2D spatial pointing
-│   └── Vantage2DPointing.tsv
+│   └── VANTAGE_2DPointing.tsv
 ├── referring/                    # 2D referring expressions
 │   └── refdrone_test_llava.json
 ├── temporal_localization/        # Temporal grounding
@@ -59,7 +59,13 @@ resolving against `event_verification/videos/`. The binary `answer` is
 removed.
 
 ### `pointing/` — 2D Spatial Pointing
-`Vantage2DPointing.tsv` — tab-separated. Each row carries the question and multiple-choice options; the last two ground-truth columns are dropped.
+`VANTAGE_2DPointing.tsv` — tab-separated, 8 columns:
+`index, question_id, image_path, question, A, B, C, D`. Each row carries
+the question and four multiple-choice options (`A`–`D`); each option is
+an `x,y` pair in the **normalized `0–1000` coordinate system** (both
+components in `[0, 1000]` relative to the image dimensions). Ground-truth
+fields (`answer`, `target_point`) are held server-side and are not
+included in the public TSV.
 
 ### `referring/` — 2D Referring Expressions
 `refdrone_test_llava.json` — list of LLaVA conversation entries. Only the `human` turn (the question) is retained; the `gpt` turn (predicted bboxes) and GT meta fields are removed.