πγβ°γTable of Contents
π Out of the many Explorers, you can be among the Shining stars who support us! βοΈ
[DepthFlow]: Image β 2.5D Parallax Effect Video. A Professional [Depthy] Alternative.
π Right click and loop me!
DepthFlow.2023.12.19.mp4
Note: Yes, the only input to DepthFlow was the Original Image
π‘ Base idea:
- Given an image and its depth map, have a shader to generate a parallax effect
- Estimate Depth Maps with Neural Networks for generic images
- Varying the projections over time generates a 2.5D video
As simple as that, we achieve a similar effect as Depthy
π΄π‘π’
Download and install our Monorepo with all the code and projects first
Note: You cannot run this project in dev mode without the Framework!
- Run the command:
broken depthflow
on the Broken Environment
A real time window should pop up. You can see broken depthflow --help
for render options
broken depthflow settings (url | path) main
broken depthflow (--render | -r)
broken depthflow -r -w 1280 -h 720 -f 30
broken depthflow -r -o ./video_name -f mkv
broken depthflow settings (url | path) main --render -s 2
Note: A high SSAA -s 2
is recommended for antialiasing
Todo: Reimplement the Gradio interface; configurable parallax settings
By default, Pytorch will be installed with CPU support.
If you want to Speed Up the Depth Estimation process, you can install it with GPU support:
NVIDIAN1 (CUDA):
*N1: Your GPU must support PyTorch's CUDA version, all >= GTX 900. Preferably install from your package manager
AMDR1 (ROCm):
- Run the command:
broken depthflow poe rocm
*R1: Check GPU compatibility, <= RX 500 series is not supported. Preferably install from your package manager
CPU / macOS (Default):
- Run the command:
broken depthflow poe cpu
See BrokenSource Repository for the License of the Code Assets, Projects and User Generated Content.