CAD_512

• Library: https://github.com/nicolas-dufour/CAD

<div align="center">

Don’t drop your samples! Coherence-aware training benefits Conditional diffusion

<a href="https://nicolas-dufour.github.io/" >Nicolas Dufour</a>, <a href="https://scholar.google.com/citations?user=n_C2h-QAAAAJ&hl=fr&oi=ao" >Victor Besnier</a>, <a href="https://vicky.kalogeiton.info/" >Vicky Kalogeiton</a>, <a href="https://davidpicard.github.io/" >David Picard</a>

</div> This repo has the code for the paper "Dont Drop Your samples: Coherence-aware training benefits Condition Diffusion" accepted at CVPR 2024 as a Highlight.

The core idea is that diffusion model is usually trained on noisy data. The usual solution is to filter massive datapools. We propose a new training method that leverages the coherence of the data to improve the training of diffusion models. We show that this method improves the quality of the generated samples on several datasets.

Project website: https://nicolas-dufour.github.io/cad

Install

To install, first create a conda env with python 3.10

conda create -n cad python=3.10

Activate the env

conda activate cad

For inference only,

pip install cad-diffusion

Pretrained models

To use the pretrained model do the following:

from cad import CADT2IPipeline

pipe = CADT2IPipeline("nicolas-dufour/CAD_512").to("cuda")

prompt = "An avocado armchair"

image = pipe(prompt, cfg=15)

If you just want to download the models, not the sampling pipeline, you can do:

from cad import CAD

model = CAD.from_pretrained("nicolas-dufour/CAD_512")

Models are hosted in the hugging face hub. The previous scripts download them automatically, but weights can be found at:

https://huggingface.co/nicolas-dufour/CAD_256

https://huggingface.co/nicolas-dufour/CAD_512

Using the Pipeline

The CADT2IPipeline class provides a comprehensive interface for generating images from text prompts. Here's a detailed guide on how to use it:

Basic Usage

from cad import CADT2IPipeline

# Initialize the pipeline
pipe = CADT2IPipeline("nicolas-dufour/CAD_512").to("cuda")

# Generate an image from a prompt
prompt = "An avocado armchair"
image = pipe(prompt, cfg=15)

Advanced Configuration

The pipeline can be initialized with several customization options:

pipe = CADT2IPipeline(
    model_path="nicolas-dufour/CAD_512",
    sampler="ddim",                    # Options: "ddim", "ddpm", "dpm", "dpm_2S", "dpm_2M"
    scheduler="sigmoid",               # Options: "sigmoid", "cosine", "linear"
    postprocessing="sd_1_5_vae",      # Options: "consistency-decoder", "sd_1_5_vae"
    scheduler_start=-3,
    scheduler_end=3,
    scheduler_tau=1.1,
    device="cuda"
)

Generation Parameters

The pipeline's __call__ method accepts various parameters to control the generation process:

image = pipe(
    cond="A beautiful landscape",          # Text prompt or list of prompts
    num_samples=4,                         # Number of images to generate
    cfg=15,                               # Classifier-free guidance scale
    guidance_type="constant",             # Type of guidance: "constant", "linear"
    guidance_start_step=0,                # Step to start guidance
    coherence_value=1.0,                  # Coherence value for sampling
    uncoherence_value=0.0,                # Uncoherence value for sampling
    thresholding_type="clamp",           # Type of thresholding: "clamp", "dynamic_thresholding", "per_channel_dynamic_thresholding"
    clamp_value=1.0,                      # Clamp value for thresholding
    thresholding_percentile=0.995         # Percentile for thresholding
)

Guidance Types

• constant: Applies uniform guidance throughout the sampling process
• linear: Linearly increases guidance strength from start to end
• exponential: Exponentially increases guidance strength from start to end

Thresholding Types

• clamp: Clamps values to a fixed range using clamp_value
• dynamic: Dynamically adjusts thresholds based on the batch statistics
• percentile: Uses percentile-based thresholding with thresholding_percentile

Advanced Parameters

For more control over the generation process, you can also specify:

• x_N: Initial noise tensor
• latents: Previous latents for continuation
• num_steps: Custom number of sampling steps
• sampler: Custom sampler function
• scheduler: Custom scheduler function
• guidance_start_step: Step to start guidance
• generator: Random number generator for reproducibility
• unconfident_prompt: Custom unconfident prompt text

Citation

If you happen to use this repo in your experiments, you can acknowledge us by citing the following paper:

@article{dufour2024dont,
  title={Don’t drop your samples! Coherence-aware training benefits Conditional diffusion},
  author={Nicolas Dufour and Victor Besnier and Vicky Kalogeiton and David Picard},
  journal={CVPR}
  year={2024}
}