Object detection project for real-time (webcam) and offline (video processing) application.
The developer of this repository has not created any items for sale yet. Need a bug fixed? Help with integration? A different license? Create a request here:
Apply tensorflow object detection on input video stream. One could use webcam (or any other device) stream or send a video file. It is possible to write Output file with detection boxes.
Requirements: Linux with docker
Clone repo in your working directory
Build docker image:
docker build -t realtime-objectdetection .
Configure script (see bellow)
Configuration is made in exec.sh at python function call:
python3 my-object-detection.py ...
All possible arguments are:
-n (--num-frames): type=int, default=0: # of frames to loop over for FPS test
-d (--display), type=int, default=0: Whether or not frames should be displayed
-f (--fullscreen), type=int, default=0: Enable full screen
-o (--output), type=int, default=0: Whether or not modified videos shall be writen
-on (--output-name), type=str, default="output": Name of the output video file
-I (--input-device), type=int, default=0: Device number input
-i (--input-videos), type=str, default="": Path to videos input, overwrite device input if used
-w (--num-workers), type=int, default=2: Number of workers
-q-size (--queue-size), type=int, default=5: Size of the queue
-l (--logger-debug), type=int, default=0: Print logger debug
python3 my-object-detection.py -d 1 -o 1
python3 my-object-detection.py -d 1 -o 1 -i test.mp4 -w 20 -q-size 150
Inputs file are in inputs/ folder
Outputs file are in outputs/ folder (.avi)
This project is intended to run on Linux. No Windows or IOS compatibility is ensured. (for IOS, it seems to be impossible to use Docker this way. See [https://apple.stackexchange.com/questions/265281/using-webcam-connected-to-macbook-inside-a-docker-container] for more information)
From Tensorflow developers: "Starting with TensorFlow 1.6, binaries use AVX instructions which may not run on older CPUs." Hence, on machines with older CPUS, one might get an SSE4.1 compatibility runtime error as discussed in Issue #14. Hence, Tensorflow pip package v1.5 or earlier needs to be used (which might lead to some other runtime error), or else Tensorflow pip package must be built from source as discussed in [https://www.tensorflow.org/install/source]