File size: 1,292 Bytes
f893548 08d61c3 683ef33 ee0ab1e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 | ---
license: mit
language:
- en
pipeline_tag: depth-estimation
tags:
- coreml
---
Based on https://github.com/fabiotosi92/ZipDepth - Converted to CoreML for use on iPhone Devices, Internal testing suggested iPhone 12> will attain good results
📊 Quantitative Results
ZipDepth achieves state-of-the-art accuracy among lightweight embedded models on NYUv2, KITTI, ETH3D, ScanNet, and DIODE, while being significantly more efficient than large pretrained models.
🏗️ Architecture
The encoder is organized in four hierarchical stages. Stages 1–2 use RepVGG reparameterizable blocks (3×3 + 1×1 + identity branches fused into a single 3×3 at inference) augmented with Strip Pooling Attention for horizontal/vertical context. Stage 3 adds Squeeze-and-Excitation channel attention and a Global Context Block. Stage 4 deepens the representation with additional RepVGG blocks.
The neck combines SPPF multi-scale pooling with a Cross-Scale Fusion module. The decoder is a lightweight FPN with a Convex Upsampling head for sub-pixel-accurate depth maps.
|