---
license: bsd-3-clause
library_name: LiteRT
pipeline_tag: object-detection
tags: [litert, tflite, on-device, android, gpu, face-detection, yunet, libfacedetection, landmarks]
base_model: ShiqiYu/libfacedetection
---

# YuNet — LiteRT (on-device face detection, fully-GPU)

[YuNet](https://github.com/ShiqiYu/libfacedetection) (ShiqiYu/libfacedetection), a tiny fast face detector
(faces + 5 landmarks), converted to **LiteRT** and running **fully on the `CompiledModel` GPU** (ML Drift) on
Android. **0.076 M params / 0.3 MB fp16.**

![YuNet — face boxes + 5 landmarks on-device LiteRT GPU](samples/sample.png)

## On-device (Pixel 8a, Tensor G3 — verified)

| | |
|---|---|
| nodes on GPU | **146 / 146** LITERT_CL (full residency) |
| inference | **~4 ms** (640×640) |
| size | **0.3 MB** (fp16) |
| accuracy | device-vs-PyTorch corr **0.9999** (all 12 outputs) |

```
image[1,3,640,640] (BGR, 0-255) →[GPU: YuNet]→ 12 outputs: cls/obj/bbox/kps × strides {8,16,32}
```

## How it converts (litert-torch) — clean, no re-authoring

Pure CNN (depthwise-separable `ConvDPUnit`) + a **nearest-upsample** neck (`F.interpolate(mode="nearest")` →
`RESIZE_NEAREST_NEIGHBOR`, no transposed conv) + non-padded `MaxPool2d` (no `PADV2`). The head's per-stride
`permute/reshape/sigmoid` is baked in → 12 decode-ready outputs. Banned ops NONE, ≤4D, tflite-vs-torch corr
**1.0**, device-vs-torch corr **0.9999**.

## Decode (host-side) & preprocessing

**Preprocessing**: letterbox to 640×640, **BGR, 0-255, no normalization**. Anchor-free priors
(`px=col·s, py=row·s`, offset 0): score=`cls·obj`, box=center+`exp(wh)·s`, 5 landmarks `kps·s+prior`, then NMS.

## License

[BSD-3-Clause](https://github.com/ShiqiYu/libfacedetection/blob/master/LICENSE). Upstream:
[ShiqiYu/libfacedetection](https://github.com/ShiqiYu/libfacedetection).