Skip to content
Deep object detection on a Raspberry Pi using Keras + Tensorflow
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
detector working demo of MobileNetv2 @ 24 fps Nov 2, 2018
examples fix model writer statement Oct 30, 2018
trainers fix num of train/validation samples in dice trainer Nov 1, 2018
utils fix num of train/validation samples in dice trainer Nov 1, 2018
.gitignore clean up repo Oct 31, 2018 reorganize directory structure to work with gcloud ml engines packager Oct 30, 2018 README first pass Nov 3, 2018 fix dev.env, add to top level Oct 30, 2018
rpi.requirements.txt use BASIC_TPU scale tier Oct 31, 2018

RPI Vision

Deep object detection on a Raspberry Pi using Tensorflow & Keras.


  • Raspberry Pi 3 Model B
  • SD card 8+ GB
  • 3.5" 480 x 320 TFT/SPI screen (XPT2046 controller)

Install Dependencies (on Raspberry Pi)

sudo apt-get update && \
sudo apt-get upgrade && \
sudo apt-get install git python3-dev python3-pip \
crossbuild-essential-armhf libatlas-base-dev   \
libhdf5-dev libhdf5-serial-dev \ 
libopenjp2-7-dev ibtiff5 build-essential cmake pkg-config && \
sudo pip3 install -U virtualenv
git clone
cd rpi-vision
pip install -r rpi.requirements.txt

Install TFT Drivers

WARNING these instructions only apply to the 3.5" TFT (XPT2046) screen. If you're using a difference size or controller, please refer to the instructions in LCD-show#README.

git clone
chmod -R 755 LCD-show
cd LCD-show
sudo ./LCD35-show

Install FBCP

This step is only neccessary if you're using an SPI Display. If you're using an HDMI display, skip this step.

Updating /boot/config.txt

For better TFT screen performance, add the following to /boot/config.txt. Refer to Raspbian's video options in config.txt if you're using a different display.

@ todo

Setup Google Cloud (optional)


Running a trainer (GPU Accelerated)

pip install -r trainer.requirements.txt

@todo API docs

Training a custom CNN

@todo API docs

Analyzing via Tensorboard

tensorboard --logdir gs://my-gcs-bucket/my-model/logs/


You can’t perform that action at this time.