Add README.md

1aa7e96 verified 2 days ago

3.41 kB

license: mit
library_name: coreml
pipeline_tag: depth-estimation
base_model: Ruicheng/moge-2-vitb-normal
base_model_relation: quantized
tags:
  - coreml
  - core-ml
  - ios
  - macos
  - apple
  - on-device
  - monocular-depth
  - metric-depth
  - surface-normals
  - dinov2
  - arxiv:2507.02546

MoGe-2 ViT-B (504×504) — Core ML

Microsoft, CVPR 2025

Monocular geometry from a single image. Metric depth, surface normals, confidence mask. DINOv2 ViT-B/14 backbone.

MoGe-2 ViT-B (504×504) demo

Core ML conversion of microsoft/MoGe for on-device inference on iPhone, iPad and Mac. Converted with coremltools; the packages are stateless, so all sequencing and buffering lives in your Swift code.


Task	depth estimation
Upstream	microsoft/MoGe
Packages	1
Download size	184 MB
Minimum iOS	17.0
Peak RAM	~600 MB

Files

File	Size	Compute units	SHA-256
`MoGe2_ViTB_Normal_504.mlpackage.zip`	184 MB	`all`	`f60cfb4804707a48…`
Total	184 MB

compute_units is not a suggestion -- it is the configuration the conversion was verified against. Moving a package to a different compute unit can silently change the numerics (FP16 attention overflow) or crash on the GPU.

Download

hf download mlboydaisuke/coreml-zoo --include "moge2/*" --local-dir ./moge2_vitb_normal_504
unzip './moge2_vitb_normal_504/moge2/*.zip' -d ./moge2_vitb_normal_504

Use in Swift

import CoreML

let config = MLModelConfiguration()
config.computeUnits = .all   // as converted — see the table above

// Unzip the .mlpackage, drop it into your Xcode target and Xcode compiles it
// at build time:
let model = try MoGe2_ViTB_Normal_504(configuration: config)

// ...or compile a downloaded .mlpackage at runtime:
let compiled = try await MLModel.compileModel(at: mlpackageURL)
let model = try MLModel(contentsOf: compiled, configuration: config)

Demo

Sample app — sample_apps/MoGe2Demo, a standalone SwiftUI project.
Models Zoo — this model is downloadable and runnable inside the Models Zoo app on the App Store, no build required.

Conversion

Script: convert_moge2.py
Pitfalls hit during conversion (FP16 overflow, ANE buffer limits, stride handling): docs/coreml_conversion_notes.md
Model index: CoreML-Models

License

The conversion inherits the upstream license: MIT.

Credits

Upstream authors: microsoft/MoGe, 2025
Core ML conversion: john-rocky (Daisuke Majima)