su5yam / pytorch-headtrip

Single Deep Dreaming and Sequence Dreaming with Optical Flow and Depth Estimation (Pytorch)

Pytorch Headtrip

• Work in Progress

Single Deep Dreaming and Sequence Dreaming with Optical Flow and Depth Estimation in Pytorch.

Check out my article for the setup!

@aertist made a colab notebook:

Features:

• Sequence Dreaming with Optical Flow (Farneback or SpyNet)
• Depth Estimation with MiDas
• Supports multiple Pytorch Architectures
• Dream Single Class of ImageNet, check my post here

Install

Requirements

1. Python 3.7
2. Pytorch 1.7
3. OpenCV
4. Matplotlib

pip install -r requirements.txt

Depending on your cuda version, you might get an error installing pytorch in your env.

Usage

For inference you need a config file like the basicconf in the _configs folder.

python dream.py --config configs/basic_conf.yaml

Currently, its only possible to dream on a sequence of frames from a video that are extracted beforehand with e.g. ffmpeg

• The SpyNet Code is adapted from this github repository

Buy me a coffee! ☕