Computer Vision Hackathon: Mini Amazon Go
Tutorial: https://docs.aws.amazon.com/kinesisvideostreams/latest/dg/producersdk-cpp-rpi.html
- The default region is
us-west-2. Change to nearest region if applicable. - Raspbian OS can use
apt-getto install shared libs to accelerate compilation of Kinesis Producer C++ SDK (https://github.com/awslabs/amazon-kinesis-video-streams-producer-sdk-cpp/blob/master/install-instructions-linux.md#install-steps-for-ubuntu-17x-and-raspbian-stretch-using-apt-get) - Setup IAM role for Kinesis Video Stream, and paste the secret key to raspberry
- There's a sample app to upload videos to KVS.
./kinesis_video_gstreamer_sample_app -w <width> -h <height> -f <framerate> -b <bitrateInKBPS> Stream Name(Note that not all resolution works, 640x480 works on Raspberry Pi 4 for me), or you can use src/start_upload.sh.
There's potential fix for large latency: https://docs.aws.amazon.com/kinesisvideostreams/latest/dg/troubleshooting.html#troubleshooting-general
For SageMaker notebook to access KVS, it must have permission AmazonKinesisVideoStreamsFullAccess added to the SageMaker Execution Role when creating the notebook instance
See VideoStream Example.
See src/training_object_detector_yolo3.ipynb and src/test_object_detector_yolo3.ipynb (Yolo3, Recommended)
Or see src/training_object_detector.ipynb and src/test_object_detector.ipynb (SSD)
Or see Beijing CV hackathon