image-diffusion
Operate 2D image DDPM/DDIM workflows with denoising-diffusion-pytorch Unet, GaussianDiffusion, Dataset, Trainer, FID, and RePaint guidance.
- License 许可
- MIT
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- image-diffusion
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Operate 2D image DDPM/DDIM workflows with denoising-diffusion-pytorch Unet, GaussianDiffusion, Dataset, Trainer, FID, and RePaint guidance.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/api-reference.md5.0 KB
- 参考文档references/troubleshooting.md3.0 KB
- 参考文档references/workflows.md4.5 KB
03 SKILL.md 正文
Image Diffusion
Use this sub-skill for 2D image diffusion tasks with denoising-diffusion-pytorch (denoising_diffusion_pytorch import), inspected at version 2.3.1.
Route here for
- Building a 2D
UnetandGaussianDiffusionfor DDPM or DDIM image loss and sampling. - Adapting the README-style image tensor recipe into a tiny CPU or CUDA smoke check.
- Using the folder
DatasetandTrainerfor image-folder training. - Explaining
calculate_fid,FIDEvaluation, andsave_best_and_latest_onlybehavior. - Using the RePaint module for mask + ground-truth image inpainting / resampling guidance.
- Debugging image shape, channel, schedule, sampling, Trainer, FID, or optional flash-attention failures.
Route elsewhere
- 1D sequence APIs (
Unet1D,GaussianDiffusion1D,Dataset1D,Trainer1D) -> ../sequence-diffusion/SKILL.md. - Classifier-free guidance, external classifier guidance, and
XMWrapper-> ../conditioning-guidance/SKILL.md. - Karras, continuous-time, EDM, simple diffusion, learned variance, weighted objective, and other advanced variants -> ../advanced-variants/SKILL.md.
Operating map
- For import paths, constructor contracts, parameter constraints, and FID/RePaint API notes, read references/api-reference.md.
- For safe setup, image loss/sample smoke, Trainer setup, DDPM vs DDIM sampling, interpolation, FID, and RePaint workflows, read references/workflows.md.
- For common failure diagnosis, read references/troubleshooting.md.
- To verify an installed package without training or downloads, run scripts/smoke_image_diffusion.py from any working directory after installing the package.
Safe defaults
Prefer tiny smoke settings before any real training: dim=8, dim_mults=(1,), channels=1, image_size=8, timesteps=8, sampling_timesteps=4, beta_schedule='sigmoid', flash_attn=False, and random tensors in [0, 1]. These defaults exercise loss and DDIM sampling without training or dataset access.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.