Depth Estimation STM32 model zoo

Depth estimation is a computer vision task that involves predicting the distance between each pixel in an image and the camera. It is a key component in applications such as robotics, autonomous navigation, and augmented reality, where understanding the 3D structure of a scene is essential.

Overview

This use case provides tools for running prediction and benchmark services on depth estimation models using the STM32AI Model Zoo framework.

A minimal set of example configuration files is available in the config_file_examples/ folder to help you get started with supported services. Additionally, the STM32AI Model Zoo repository offers ready-to-use models and their configuration .yaml files.

How to Use

To run the services, use the main launcher script stm32ai_main.py with a configuration file (e.g. user_config.yaml).

For general instructions on writing and customizing your YAML files, please refer to the README_BENCHMARKING.md and README_PREDICTION.md.

Supported Services

Service	Description
predict	Runs inference on a single image or a folder of images using a depth model.
benchmark	Evaluates model performance (e.g. speed, memory, and optionally accuracy).

Other services like training, quantization, or evaluation are not currently available for this use case.