|
| 1 | +<!--Copyright 2022 The HuggingFace Team. All rights reserved. |
| 2 | + |
| 3 | +Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with |
| 4 | +the License. You may obtain a copy of the License at |
| 5 | + |
| 6 | +http://www.apache.org/licenses/LICENSE-2.0 |
| 7 | + |
| 8 | +Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on |
| 9 | +an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the |
| 10 | +specific language governing permissions and limitations under the License. |
| 11 | +--> |
| 12 | + |
| 13 | +# Cycle Diffusion |
| 14 | + |
| 15 | +## Overview |
| 16 | + |
| 17 | +Cycle Diffusion is a Text-Guided Image-to-Image Generation model proposed in [Unifying Diffusion Models' Latent Space, with Applications to CycleDiffusion and Guidance](https://arxiv.org/abs/2210.05559) by Chen Henry Wu, Fernando De la Torre. |
| 18 | + |
| 19 | +The abstract of the paper is the following: |
| 20 | + |
| 21 | +*Diffusion models have achieved unprecedented performance in generative modeling. The commonly-adopted formulation of the latent code of diffusion models is a sequence of gradually denoised samples, as opposed to the simpler (e.g., Gaussian) latent space of GANs, VAEs, and normalizing flows. This paper provides an alternative, Gaussian formulation of the latent space of various diffusion models, as well as an invertible DPM-Encoder that maps images into the latent space. While our formulation is purely based on the definition of diffusion models, we demonstrate several intriguing consequences. (1) Empirically, we observe that a common latent space emerges from two diffusion models trained independently on related domains. In light of this finding, we propose CycleDiffusion, which uses DPM-Encoder for unpaired image-to-image translation. Furthermore, applying CycleDiffusion to text-to-image diffusion models, we show that large-scale text-to-image diffusion models can be used as zero-shot image-to-image editors. (2) One can guide pre-trained diffusion models and GANs by controlling the latent codes in a unified, plug-and-play formulation based on energy-based models. Using the CLIP model and a face recognition model as guidance, we demonstrate that diffusion models have better coverage of low-density sub-populations and individuals than GANs.* |
| 22 | + |
| 23 | +*Tips*: |
| 24 | +- The Cycle Diffusion pipeline is fully compatible with any [Stable Diffusion](./stable_diffusion) checkpoints |
| 25 | +- Currently Cycle Diffusion only works with the [`DDIMScheduler`]. |
| 26 | + |
| 27 | +*Example*: |
| 28 | + |
| 29 | +In the following we should how to best use the [`CycleDiffusionPipeline`] |
| 30 | + |
| 31 | +```python |
| 32 | +import requests |
| 33 | +import torch |
| 34 | +from PIL import Image |
| 35 | +from io import BytesIO |
| 36 | + |
| 37 | +from diffusers import CycleDiffusionPipeline, DDIMScheduler |
| 38 | + |
| 39 | +# load the pipeline |
| 40 | +# make sure you're logged in with `huggingface-cli login` |
| 41 | +model_id_or_path = "CompVis/stable-diffusion-v1-4" |
| 42 | +scheduler = DDIMScheduler.from_config(model_id_or_path, subfolder="scheduler") |
| 43 | +pipe = CycleDiffusionPipeline.from_pretrained(model_id_or_path, scheduler=scheduler).to("cuda") |
| 44 | + |
| 45 | +# let's download an initial image |
| 46 | +url = "https://raw.githubusercontent.com/ChenWu98/cycle-diffusion/main/data/dalle2/An%20astronaut%20riding%20a%20horse.png" |
| 47 | +response = requests.get(url) |
| 48 | +init_image = Image.open(BytesIO(response.content)).convert("RGB") |
| 49 | +init_image = init_image.resize((512, 512)) |
| 50 | +init_image.save("horse.png") |
| 51 | + |
| 52 | +# let's specify a prompt |
| 53 | +source_prompt = "An astronaut riding a horse" |
| 54 | +prompt = "An astronaut riding an elephant" |
| 55 | + |
| 56 | +# call the pipeline |
| 57 | +image = pipe( |
| 58 | + prompt=prompt, |
| 59 | + source_prompt=source_prompt, |
| 60 | + init_image=init_image, |
| 61 | + num_inference_steps=100, |
| 62 | + eta=0.1, |
| 63 | + strength=0.8, |
| 64 | + guidance_scale=2, |
| 65 | + source_guidance_scale=1, |
| 66 | +).images[0] |
| 67 | + |
| 68 | +image.save("horse_to_elephant.png") |
| 69 | + |
| 70 | +# let's try another example |
| 71 | +# See more samples at the original repo: https://github.com/ChenWu98/cycle-diffusion |
| 72 | +url = "https://raw.githubusercontent.com/ChenWu98/cycle-diffusion/main/data/dalle2/A%20black%20colored%20car.png" |
| 73 | +response = requests.get(url) |
| 74 | +init_image = Image.open(BytesIO(response.content)).convert("RGB") |
| 75 | +init_image = init_image.resize((512, 512)) |
| 76 | +init_image.save("black.png") |
| 77 | + |
| 78 | +source_prompt = "A black colored car" |
| 79 | +prompt = "A blue colored car" |
| 80 | + |
| 81 | +# call the pipeline |
| 82 | +torch.manual_seed(0) |
| 83 | +image = pipe( |
| 84 | + prompt=prompt, |
| 85 | + source_prompt=source_prompt, |
| 86 | + init_image=init_image, |
| 87 | + num_inference_steps=100, |
| 88 | + eta=0.1, |
| 89 | + strength=0.85, |
| 90 | + guidance_scale=3, |
| 91 | + source_guidance_scale=1, |
| 92 | +).images[0] |
| 93 | + |
| 94 | +image.save("black_to_blue.png") |
| 95 | +``` |
| 96 | + |
| 97 | +## CycleDiffusionPipeline |
| 98 | +[[autodoc]] CycleDiffusionPipeline |
| 99 | + - __call__ |
0 commit comments