Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,37 @@
+---
+tags:
+- time-series
+- kinematics
+- biomechanics
+- gait-analysis
+- transformer
+license: apache-2.0
+library_name: pytorch
+pipeline_tag: time-series-classification
+---
+
+# PoseT5: Compact Multivariate Gait Kinematics & Keystroke Predictor
+
+PoseT5 is a lightweight (~430K parameter) time-series transformer architecture designed for real-time 3D skeletal motion capture and inverse kinematic keystroke estimation.
+
+## Organization
+Trained and hosted under **[Learningbase](https://huggingface.co/Learningbase)**.
+
+## Performance Metrics
+- **Validation Macro F1**: 0.7110
+- **Hamming Loss**: 0.2041
+
+| Keystroke | Action | F1 Score | Precision | Recall |
+|---|---|---|---|---|
+| `W` | Forward Walk/Run | 0.940 | 0.940 | 0.940 |
+| `A` | Yaw Turn Left | 0.385 | 0.263 | 0.724 |
+| `S` | Backward Move | 0.947 | 0.900 | 1.000 |
+| `D` | Yaw Turn Right | 0.440 | 0.289 | 0.923 |
+| `Space` | Jump Dynamics | 0.842 | 0.800 | 0.889 |
+
+## Architecture Details
+- **Input Channels**: 22 humanoid anatomical joints $\times$ 7 continuous DOF = **154 continuous channels**.
+- **Instance Normalization**: Reversible Instance Normalization (RevIN).
+- **Temporal Patching**: Slices 32 frames into 4 patches ($P=8$) projected into $d_{\text{model}} = 128$.
+- **Backbone**: 2-layer Multi-Head Self-Attention Transformer Encoder ($430\text{K}$ parameters).
+- **Efficiency**: Sub-millisecond CPU inference latency (< 2ms).