This is a modified version of Darknet.keras, with some bugs fixed and YOLO Model support.
###Contributions
- Modified DarkNet.py to load YOLO Model(rather than YOLO Tiny Model) weights files.
- Modified RunTinyYOLO.py, which had a few bugs, to reproduce YOLO detection results in darknet.
###TODO
- Building network structure by hand is tedious, should write code to do so automatically from cfg files
- Implement training code
###The original README This is a transfer from weights trained by Darknet to keras with Theano backend.
Currently I only finish test process for YOLO Tiny Model.
Due to image preprocess difference bewteen my reimplementaion and Darknet's, the result has slightly difference.
To run the code:
- Create three folders:
weights, results, images
. - Put the images you want to detect in the folder
images
. - Download yolo-tiny.weights from Darknet website and put it in the folder
weights
. - Run:
python RunTinyYolo.py
The detection result will be saved in the results folder
Notes: Make sure your have Theano and Keras installed