mlboydaisuke/Depth-Anything-3-Base-CoreML

Depth Anything 3 Base (504×504) — Core ML

ByteDance-Seed, ICLR 2026 oral

Relative monocular depth from a single image. DA3 Main Series, Base (0.12B params, DINOv2 ViT-B/14 + DualDPT head). Higher quality than Small at ~3× the model size.

Core ML conversion of ByteDance-Seed/Depth-Anything-3 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	ByteDance-Seed/Depth-Anything-3
Packages	1
Download size	173 MB
Minimum iOS	17.0
Peak RAM	~700 MB

Files

File	Size	Compute units	SHA-256
DepthAnythingV3_base_504.mlpackage.zip	173 MB	all	cd96d12b7d14fb92…
Total	173 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 "depth_anything_v3/*" --local-dir ./depth_anything_v3_base_504
unzip './depth_anything_v3_base_504/depth_anything_v3/*.zip' -d ./depth_anything_v3_base_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 DepthAnythingV3_base_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

• Models Zoo — this model is downloadable and runnable inside the Models Zoo app on the App Store, no build required.

Conversion

• Script: convert_depth_anything_v3.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: Apache-2.0.

Credits

• Upstream authors: ByteDance-Seed/Depth-Anything-3, 2025
• Core ML conversion: john-rocky (Daisuke Majima)