This repository contains Tensorflow code to create, train, and deploy a monocular depth estimation model for an STM32 microcontroller.
pip install -r requirements.txt- create a .env file in the project root with the following variables:
path_to_project_dir=/abs/path/to/project/root
base_dataset_dir=/abs/path/to/nyuv2/datasetPlease, replace absolute paths where appropriate. For interfacing with a device, set a proper serial port that your device is using in depth_inference.py
full_pipeline.ipynb is a notebook that trains models with multiple quantization and pruning techniques, storing them under models/ and cfiles/ folders relative to the project root.
Trained models will appear in the models/ folder. .h5 or .tflite files can be used within STM32 to profile the model and validate its performance in an environment similar to the real MCU.
cfiles/ contains C files for each model type. Flashing a model to a device requires only its corresponding header file in cfiles/.
depth_inference.py is used for interfacing with a device and prediction.png is a sample prediction obtained after running the script.