Yolo (You only look once, Unifed, Real Time Object Detection) implementated in Python 3 with Tensorflow. Based on a paper by Joseph Redmon (https://pjreddie.com/media/files/papers/yolo_1.pdf)
Install the following : Python 3, Tensorflow, Opencv
If on linux : In terminal, in the folder yolo run this
chmod u+x download_pascal.sh download_weights
./download_weights
If on Windows: Download the weights with one of the links, extract the file and put the result in the folder and put them network\data\weights folder.
Link : https://www.dropbox.com/s/oit8og2t2on0j5t/weights_JG.zip?dl=0 or https://www.dropbox.com/s/7uyw95l2qgebvoe/weights.tar
On linux, before the first training session, run this command in terminal in the folder yolo
./download_pascal.sh
The file training.py trains the network.
python training.py -gpu
"-gpu" is a flag to use the gpu if one is present
Before the first time, only run the weights shell scripts
The file inference.py does inference on images and video.
python inference.py -test_img -gpu
- Flags
- "-gpu" : use the gpu if one is present
- "-test_img" and "-test_video" : run the network on files in the test folder
- "-file_path FILE_PATH" : give the path to a file and the network will process that file
- "-save" : saves the results
- "-h" : gives information on the flags
Cats
Dog and boy
Dog and boy
-
Tensorflow - An open-source software library for Machine Intelligence
-
OpenCV - OpenCV is a library of programming functions mainly aimed at real-time computer vision
- Jean-Gabriel Simard