This tool helps to create animations with deepdream. Extract frames from videos, process them with deepdream and then output as new video file. Frame blending option is provided, to ensure "stable" dreams across frames. A preview function make rapid iterations possible. Optionally guided dreaming & optical flow can be used.
-
Extract Video
python dreamer.py --input myvideo/video.mp4 --output myvideo --extract 1 -
Run DeepDream
python dreamer.py --input myvideo --output myvideo/frames -
Create Video
python dreamer.py --input myvideo/frames --output myvideo/deepdreamvideo.mp4 --create 1
(change "myvideo" to your directory/file name)
(change the path of the caffe model inside dreamer.py to where your model is )
Use Optical Flow & Guided Dreams & GPU
python dreamer.py --input myvideo --output myvideo/frames --octaves 4 --octavescale 1 --iterations 10 --jitter 32 --zoom 1 --stepsize 1.5 --blend 0 --layers inception_3b/output --gpu 1 --flow 1 --guide guide/flowers.png
Create a preview (currently does not work with flow)
python dreamer.py --input myvideo --output myvideo/frames --preview 600
Tweak settings
python dreamer.py --input myvideo --output myvideo/frames --preview 600 --octaves 4 --octavescale 1.4 --iterations 10 --jitter 32 --zoom 1 --stepsize 1.5 --blend 0.5 --layers inception_3a/output inception_3b/output
(Preview changes Images to width of choice, original files not changed)
(Try using multiple layers, it will cycle through them from frame to frame.)
Use the above commands and stack them by putting a ";" inbetween commands.
python dreamer.py --input myvideo --output myvideo/frames;python dreamer.py --input myvideo2 --output myvideo2/frames
DeepDream Music Video
Optical flow + Guided Dreaming tests
It would be very helpful for other deepdream researchers, if you could include the used parameters in the description of your youtube videos.
- Python
- Caffe (and other deepdream dependencies)
- FFMPEG
- CV2 (if you use optical flow)
The tool currently does not handle audio. Check out DeepDreamVideo by @graphific
