Skip to content

Latest commit

 

History

History
 
 

gen2-depth-mbnv2

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
README.md
README.md
 
 
main.py
main.py
 
 
requirements.txt
requirements.txt
 
 

README.md

[Gen2] Monocular Depth Estimation \w Transfer Learning on DepthAI

This example shows an implementation of Monocular Depth Estimation with Transfer Learning pretrained MobileNetV2 on DepthAI in the Gen2 API system. Blob is created with ONNX from PINTO's Model ZOO, which is then converted to OpenVINO IR with required flags and converted to blob.

There are two available blob's for different input sizes:

  • depth_estimation_mbnv2_240x320_openvino_2021.4_6shave, for 240x320 (H x W) input, which runs with ~15 FPS, and
  • depth_estimation_mbnv2_480x640_openvino_2021.4_6shave, for 480x640 (H x W) input, which runs with ~5 FPS.

Output of the model is a unnormalized density map of size (INPUT_HEIGHT / 2) x (INPUT_WIDTH / 2). We rescale the input frame to have the same size when showing the demo.

Image example

Pre-requisites

Install requirements:

python3 -m pip install -r requirements.txt

Usage

python3 main.py [options]

Options:

  • -w, --width: Select width of the model for inference. Default: 320. Possible options: 320 or 640.