Add CVPR 2026 banner + 4 trace overlay GIFs + cleaner README

README.md

@@ -1,5 +1,5 @@
 ---
-pretty_name: SemanticVLA TraceX 240K Fractal
+pretty_name: SemanticVLA TraceX 240K Fractal (RT-1)
 tags:
 - robotics
 - lerobot
@@ -16,48 +16,84 @@ configs:
     path: data/**/*.parquet
 ---
 
-# SemanticVLA TraceX 240K Fractal
+# SemanticVLA TraceX 240K · Fractal (RT-1)
 
-LeRobot v3.0 repack of the Fractal component in the SemanticVLA TraceX 240K release. This
-package preserves the component episode order and raw observation/action fields
-from the full-preserve v2-style release package, while reducing the file count
-by grouping multiple episodes per parquet/video file.
+> :tada: **Accepted to CVPR 2026.**
+> :black_nib: Fei Ni¹, Zhuo Chen², Yifu Yuan³, Zibin Dong³, Xianze Yao³, Shan Luo², Jianye Hao³, Jiankang Deng¹†, Stefanos Zafeiriou¹†     :school: ¹Imperial College London   ²King's College London   ³Tianjin University     :e-mail: [f.ni@imperial.ac.uk](mailto:f.ni@imperial.ac.uk)
 
-Trace columns in this v3 package:
+The **Fractal (RT-1)** component of [**TraceX-240K**](https://hf.co/collections/spikefly/semanticvla-datasets) — the trace-annotated trajectory corpus introduced in **SemanticVLA**. This package is a LeRobot v3.0 repack of Fractal · Google Robot · Open-X-Embodiment fractal20220817 v0.1.0 with **dense per-frame end-effector trace labels** stored directly in every frame row.
 
-- `trace.x`: `float32[1]`, normalized image-space x coordinate on a 0-100 scale.
-- `trace.y`: `float32[1]`, normalized image-space y coordinate on a 0-100 scale.
+> :books: **Code · checkpoints · paper:** https://github.com/Fei-Ni/SemanticVLA_Offcial
 
-The v2-style `observation.trace.xy` column is intentionally not present in this
-v3 experiment.
+## :movie_camera: Sample trace overlays
 
-## Contents
+Each clip shows the orange trace label (per-frame gripper position) overlaid on the original episode camera feed.
 
-- Repository: `spikefly/SemanticVLA-TraceX-240K-Fractal`
-- Format: LeRobot v3.0
-- Episodes: 87,182
-- Frames: 3,786,274
-- FPS: 5
-- Robot type: `google_robot`
-- Source dataset: Open-X-Embodiment fractal20220817_data v0.1.0
-- Data file target size: 100 MB
-- Video file target size: 500 MB
+<table>
+  <tr>
+    <td><img src="assets/trace_overlays/ep_017431.gif" width="200" alt="episode 17431"></td>
+    <td><img src="assets/trace_overlays/ep_034901.gif" width="200" alt="episode 34901"></td>
+    <td><img src="assets/trace_overlays/ep_043630.gif" width="200" alt="episode 43630"></td>
+    <td><img src="assets/trace_overlays/ep_061060.gif" width="200" alt="episode 61060"></td>
+  </tr>
+  <tr>
+    <td align="center"><sub>episode 17431</sub></td><td align="center"><sub>episode 34901</sub></td><td align="center"><sub>episode 43630</sub></td><td align="center"><sub>episode 61060</sub></td>
+  </tr>
+</table>
 
-## Video Streams
+## :compass: Trace label schema
+
+Dense per-frame end-effector position columns, stored directly in every frame row:
+
+- `trace.x` — `float32[1]`, normalized image-space x coordinate on a `[0, 100]` scale.
+- `trace.y` — `float32[1]`, normalized image-space y coordinate on a `[0, 100]` scale.
+
+`(0, 0)` is the top-left corner of the frame, `(100, 100)` the bottom-right.
+
+## :package: Contents
+
+| Field | Value |
+|---|---|
+| HF repo | `spikefly/SemanticVLA-TraceX-240K-Fractal` |
+| Format | LeRobot v3.0 |
+| Episodes | 87,182 |
+| Frames | 3,786,274 |
+| FPS | 5 |
+| Robot type | `google_robot` |
+| Source dataset | Open-X-Embodiment fractal20220817_data v0.1.0 |
+| Data file target | 100 MB |
+| Video file target | 500 MB |
+
+### Video streams
 
 - `observation.images.image`
 
-## Coverage
+## :compass: Coverage
+
+The upstream Fractal train split has 87,212 episodes. This TraceX v3 package contains the 87,182 episodes with valid dense trace annotations; the 30 episodes without dense trace are excluded (no synthetic fallback inserted).
+
+## :open_file_folder: Sibling datasets in TraceX-240K
+
+<table>
+  <tr><th>Dataset</th><th>HF Repository</th><th>Embodiment</th></tr>
+  <tr><td><b>Bridge</b></td><td><a href="https://huggingface.co/datasets/spikefly/SemanticVLA-TraceX-240K-Bridge">spikefly/SemanticVLA-TraceX-240K-Bridge</a></td><td>BridgeData V2 / WidowX</td></tr>
+  <tr><td><b>BC-Z</b></td><td><a href="https://huggingface.co/datasets/spikefly/SemanticVLA-TraceX-240K-BC-Z">spikefly/SemanticVLA-TraceX-240K-BC-Z</a></td><td>BC-Z / Google Robot</td></tr>
+  <tr><td><b>DROID</b></td><td><a href="https://huggingface.co/datasets/spikefly/SemanticVLA-TraceX-240K-DROID">spikefly/SemanticVLA-TraceX-240K-DROID</a></td><td>DROID / Franka</td></tr>
+</table>
 
-The original Fractal train split has 87,212 episodes. This TraceX v3 package contains the 87,182 episodes with valid dense trace annotations. The 30 episodes without dense trace are excluded; no synthetic fallback trace is inserted.
+## :pencil: Citation
 
+If you use this dataset, please cite SemanticVLA:
 
-## Notes
+```bibtex
+@inproceedings{ni2026semanticvla,
+  title     = {SemanticVLA: Towards Semantic Reasoning over Action Memorization via Synergistic Explicit Trace and Latent Action Planning},
+  author    = {Ni, Fei and Chen, Zhuo and Yuan, Yifu and Dong, Zibin and Yao, Xianze and Luo, Shan and Hao, Jianye and Deng, Jiankang and Zafeiriou, Stefanos},
+  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
+  year      = {2026}
+}
+```
 
-The `configs` block in this card is intentional: it makes Hugging Face Data
-Studio read `data/**/*.parquet` as the train split, so the parquet columns such
-as `action`, `observation.*`, `trace.x`, and `trace.y` are visible instead of
-auto-detecting the repository as a video-only dataset.
+## :scroll: License & terms
 
-Use of this dataset is subject to the original source dataset terms, plus the
-terms of the added SemanticVLA TraceX annotations.
+The trace annotation labels (`trace.x` / `trace.y`) are released under the [MIT License](https://github.com/Fei-Ni/SemanticVLA_Offcial/blob/main/LICENSE) as part of the SemanticVLA code release. Use of the underlying robot trajectory data is subject to the **original source dataset terms** (Open-X-Embodiment fractal20220817_data v0.1.0). Please consult the upstream dataset license before use.