Add dataset card

README.md

@@ -0,0 +1,70 @@
+---
+license: cc-by-4.0
+language:
+- en
+tags:
+- video-generation
+- image-to-video
+- visual-reasoning
+- benchmark
+pretty_name: VBVR-Pro-Bench
+size_categories:
+- n<1K
+---
+
+# VBVR-Pro-Bench
+
+The frozen **evaluation split** of VBVR-Pro. 100 parameterized reasoning tasks, 5 held-out instances each, in both a video (I2V) and an interleaved-image setting.
+
+Part of the **VBVR (Very Big Video Reasoning Suite)** project: <https://video-reason.com>.
+
+Companion resources: [VBVR-Pro-SFT-Video](https://huggingface.co/datasets/Video-Reason/VBVR-Pro-SFT-Video) · [VBVR-Pro-SFT-Image](https://huggingface.co/datasets/Video-Reason/VBVR-Pro-SFT-Image) · [VBVR-Pro-RL](https://huggingface.co/datasets/Video-Reason/VBVR-Pro-RL)
+
+## At a glance
+
+| Property | Value |
+|---|---|
+| Tasks | **100** (`In-Domain_50` 50 + `Out-of-Domain_50` 50) |
+| Instances per task | **5** |
+| Total instances | **500** per setting (1,000 across both) |
+| Settings | Video (I2V) and Image (interleaved generation) |
+| Resolution | **1024 × 1024** |
+| Video | 16 fps, MPEG-4 Part 2, as generated (no re-encode) |
+| Size | 83.9 MB (video) + 9.8 MB (image) |
+
+`In-Domain_50` covers task families that also appear in the training splits; `Out-of-Domain_50` covers task families that do **not**, and is the split to report generalization on.
+
+## Layout
+
+```
+.
+├── VBVR-Pro-Bench-Video.tar.gz
+└── VBVR-Pro-Bench-Image.tar.gz
+```
+
+Extracting either archive yields:
+
+```
+VBVR-Pro-Bench-<Video|Image>/
+└── <In-Domain_50|Out-of-Domain_50>/
+    └── <TASK>/                       # e.g. G-131_select_next_figure_increasing_size_sequence_data-generator
+        └── 00000/ … 00004/
+```
+
+Per instance:
+
+| Setting | Files |
+|---|---|
+| Video | `first_frame.png`, `ground_truth.mp4`, `final_frame.png`, `prompt.txt` |
+| Image | `first_frame.png`, `frame_1.png` … `frame_N.png`, `prompt.txt` |
+
+`first_frame.png` is the conditioning image, `prompt.txt` the instruction, and the remaining files the reference output. `N` varies by task (1–21) according to how many steps the instruction asks the model to render.
+
+## Usage
+
+```bash
+huggingface-cli download Video-Reason/VBVR-Pro-Bench --repo-type dataset --local-dir ./VBVR-Pro-Bench
+cd VBVR-Pro-Bench && tar xzf VBVR-Pro-Bench-Video.tar.gz
+```
+
+Generate one video (or image sequence) per instance from `first_frame.png` + `prompt.txt`, then score with [VBVR-EvalKit](https://github.com/Video-Reason/VBVR-EvalKit).