DepthEstimation

Depth Estimation (ailia SDK for Unity)

This folder provides a monocular depth estimation sample using ailia SDK. The scene is DepthEstimation/DepthEstimationSample.unity, driven by DepthEstimation/AiliaDepthEstimatorsSample.cs.

Supported AI Models

• MiDaS
  • Purpose: Predict dense relative depth from a single RGB image.
  • Source: ./AiliaDepthEstimatorsSample.cs
  • Script behavior: Downloads midas.onnx (+ .onnx.prototxt), opens it via AiliaModel, and feeds a preprocessed camera/frame image. Uses a compute shader (optional) or CPU to prepare ImageNet‑style normalized input. Postprocess scales and normalizes output to a grayscale depth map and renders it to a RawImage.

Model Files and Paths

• midas.onnx (+ midas.onnx.prototxt) — downloaded to Application.temporaryCachePath via AiliaDownload.

Reusing in Your Own Code

• Create and open the model:
  • var model = new AiliaModel();
  • model.OpenFile(pathToPrototxt, pathToOnnx);
• Input preparation:
  • Resize your image to the model input shape from model.GetInputShape().
  • Normalize as in InputDataProcessingCPU/InputDataProcessingPSP (ImageNet mean/std, channel‑first when required).
• Run inference and visualize:
  • model.Predict(output, input);
  • Map output to 0–255 and draw to a texture (see LabelPaintMidas).

Source Links

• Sample controller: ./AiliaDepthEstimatorsSample.cs