TensorFlow implementation of estimating image aesthetics
Will people like your image
Katharina Schwarz Patrick Wieschollek, Hendrik P.A. Lensch
WACV 2018
The original model was trained on an out-dated TensorFlow training interface. Hre, we provide a multi-GPU training script based on TensorPack. For more details, see the training directory.
We provide an example to do video spotting. The steps are:
user@host $ VIDEO=original.mp4
user@host $ avconv -i original.mp4 -s 256x256 -strict -2 echo ${VIDEO%.mp4}-small.mp4 # resize the video to match the network input
user@host $ python apply.py --mp4 ${VIDEO%.mp4}-small.mp4 # apply our neural network
user@host $ python gui.py # use the GUI to display the video and the scores
Click Load Video
and select the original high-res video.
The file arod/list.txt
provides the url to images used during training. The format is url;faves;views
.