Real-time object detection for disaster response using Transfer Learning Toolkit

A notebook that demonstrates how to use the NVIDIA Intelligent Video Analytics suite to detect objects in real-time. We use Transfer Learning Toolkit to train a fast and accurate detector and DeepStream to run that detector on an NVIDIA Jetson edge device.

Getting Started Guide

Sign up for a free NVIDIA GPU Cloud (NGC) account at ngc.nvidia.com, then pull the Transfer Learning Toolkit (TLT) container.

DATA_DIR=/path/to/your/data
WORKING_DIR=/path/to/workingdir # include the "specs" and "deepstream" directories

docker pull nvcr.io/nvidia/tlt-streamanalytics:v1.0_py2
docker run --runtime=nvidia -it -v $DATA_DIR:/data \ 
        -v $WORKING_DIR:/src -p 8888:8888 \ 
        nvcr.io/nvidia/tlt-streamanalytics:v1.0_py2 /bin/bash

Launch jupyter notebook

cd /src
jupyter notebook --ip=0.0.0.0 --allow-root

Now navigate in your browser (we use Chrome) to <your host IP>:8888. If working on a local machine you can go to localhost:8888.

References

This notebook accompanies the Real-time object detection for disaster response using Transfer Learning Toolkit webinar, available at info.nvidia.com/real-time-object-detection-for-disaster-response-reg-page.html.

For more information on Transfer Learning Toolkit: developer.nvidia.com/transfer-learning-toolkit