This repo contains the unofficial implementation for Hierarchy Flow For High-Fidelity Image-to-Image Translation Fan et al. (2023), which I developed as a nice entry into the world of normalizing flows.

The authors release their official implementation which can be found here.

import torch
from src.hflow import HierarchicalFlow

flow = HierarchyFlow(
    inp_channels=3,
    out_channels=[30, 120],
    pad_size=10,
)

x = torch.randn(1, 3, 256, 256)
y = flow(x)