Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- pose-estimation
|
| 5 |
+
- vitpose
|
| 6 |
+
- mlx
|
| 7 |
+
- mlx-swift
|
| 8 |
+
- on-device
|
| 9 |
+
- apple-silicon
|
| 10 |
+
- keypoint-detection
|
| 11 |
+
library_name: mlx
|
| 12 |
+
base_model: usyd-community/vitpose-base-simple
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
# ViTPose base-simple — MLX
|
| 16 |
+
|
| 17 |
+
ViTPose (`vitpose-base-simple`) converted to **MLX** for on-device human pose
|
| 18 |
+
estimation on Apple Silicon. Weights are float16.
|
| 19 |
+
|
| 20 |
+
Built for **[MLXPose](https://github.com/NazarKozak/MLXPose)** — a native MLX Swift
|
| 21 |
+
ViTPose implementation. The Swift forward pass is numerically verified against the
|
| 22 |
+
Hugging Face reference (heatmaps `max|Δ|=1.5e-6`, decoded keypoints `max 3e-5 px`).
|
| 23 |
+
|
| 24 |
+
- Backbone: plain ViT-base (12 layers, dim 768), patch 16, input 256×192.
|
| 25 |
+
- Head: simple decoder → 17 COCO keypoint heatmaps (64×48).
|
| 26 |
+
- Conversion: [`convert_vitpose_to_mlx.py`](https://github.com/NazarKozak/MLXPose/blob/main/scripts/convert_vitpose_to_mlx.py).
|
| 27 |
+
|
| 28 |
+
## Files
|
| 29 |
+
- `weights.safetensors` — MLX float16 weights.
|
| 30 |
+
- `config.json` — original ViTPose config.
|
| 31 |
+
|
| 32 |
+
## License
|
| 33 |
+
Apache-2.0. Pretrained weights derive from COCO/MPII training data — review dataset
|
| 34 |
+
terms for your use case.
|