Just note that I used the classical ssd_mobilenet_v2_coco model from Tensorflow for speed performance. Instead, I will show how I use Docker in my all-day jobs as data scientist.

I will not spend time describing Tensorflow object-detection API implementation, since there is ton of articles on this subject. In addition, I added a video post-processing feature to my project also using multiprocessing to reduce processing time (which could be very very long when using raw Tensorflow object detection API).īoth real-time and video processing can run with high performances on my personal laptop using only 8GB CPU. Main difficulty here was to deal with video stream going into and coming from the container. To go further and in order to enhance portability, I wanted to integrate my project into a Docker container.

I started from this excellent Dat Tran article to explore the real-time object detection challenge, leading me to study python multiprocessing library to increase FPS with the Adrian Rosebrock’s website. Video processing test with Youtube video Motivation

Tensorboard docker on mac