Upload README.md with huggingface_hub

README.md

@@ -1,28 +1,64 @@
-# SimMotion-Real Benchmark (SemanticMoments)
+---
+license: mit
+task_categories:
+  - video-classification
+  - feature-extraction
+tags:
+  - video
+  - motion
+  - similarity
+  - retrieval
+  - benchmark
+language:
+  - en
+pretty_name: SimMotion-Real
+size_categories:
+  - n<1K
+---
 
-This folder contains the real-world benchmark for evaluating motion representation consistency as described in:
+# SimMotion-Real Benchmark
 
-**"SemanticMoments: Training-Free Motion Similarity via Third Moment Features"**
+Real-world benchmark for evaluating motion representation consistency, introduced in:
+
+**"SemanticMoments: Training-Free Motion Similarity via Third Moment Features"** ([arXiv:2602.09146](https://arxiv.org/abs/2602.09146))
+
+## Dataset Description
 
-## Dataset Structure
 The benchmark consists of **40 real-world test cases**, each organized as a triplet:
-1. **`ref.mp4`**: The reference video defining the target semantic motion.
-2. **`positive.mp4`**: A real-world video sharing the same semantic motion as the reference.
-3. **`negative.mp4`**: **Hard Negative**. A video with a similar visual appearance to the reference, but containing a different motion.
+
+| File | Description |
+|------|-------------|
+| `ref.mp4` | Reference video defining the target semantic motion |
+| `positive.mp4` | Video sharing the same semantic motion as reference |
+| `negative.mp4` | Hard negative - similar appearance but different motion |
+
+## Usage
+
+```python
+from semantic_moments import SimMotionReal, download_simmotion
+
+# Download
+download_simmotion(dataset="real")
+
+# Load
+dataset = SimMotionReal("SimMotion_Real_benchmark")
+print(f"Loaded {len(dataset)} triplets")
+
+for triplet in dataset:
+    print(triplet.ref_path, triplet.positive_path, triplet.negative_path)
+```
+
+Or download directly:
+```bash
+huggingface-cli download Shuberman/SimMotion-Real --repo-type dataset --local-dir SimMotion_Real_benchmark
+```
 
 ## Evaluation Protocol
-This benchmark evaluates retrieval performance using a large-scale distractor pool to ensure motion representations are robust against diverse real-world actions.
 
-* **Retrieval Pool**: For each reference video, the candidate pool includes:
-    * The corresponding `positive.mp4`.
-    * The corresponding `hard negative.mp4`.
-    * **1,000 Distractor Videos**: The first 1,000 videos from the Kinetics-400 validation set (sorted alphabetically).
-* **Metric**: Top-1 Accuracy. A trial is successful if:
-$$Similarity(ref, positive) > Similarity(ref, candidates_{all})$$
-(Where $candidates_{all}$ includes the hard negative and all 1,000 Kinetics distractors).
+- **Retrieval Pool**: For each reference, candidates include the positive, hard negative, and 1,000 Kinetics-400 distractors
+- **Metric**: Top-1 Accuracy - success if positive is retrieved first
 
 ## Citation
-If you use this benchmark, please cite:
 
 ```bibtex
 @article{huberman2026semanticmoments,
@@ -31,4 +67,8 @@ If you use this benchmark, please cite:
   journal={arXiv preprint arXiv:2602.09146},
   year={2026}
 }
-```
+```
+
+## License
+
+MIT