https://github.com/leffff/flowmodels

The aim of this repository is to test and implement Flow-Matching-based models
https://github.com/leffff/flowmodels

diffusion flow-matching inverse-problems

Last synced: 5 months ago
JSON representation

The aim of this repository is to test and implement Flow-Matching-based models

• Host: GitHub
• URL: https://github.com/leffff/flowmodels
• Owner: leffff
• License: mit
• Created: 2024-05-17T18:56:28.000Z (about 2 years ago)
• Default Branch: main
• Last Pushed: 2025-01-15T11:10:53.000Z (over 1 year ago)
• Last Synced: 2025-01-15T13:12:12.844Z (over 1 year ago)
• Topics: diffusion, flow-matching, inverse-problems
• Language: Jupyter Notebook
• Homepage:
• Size: 68.2 MB
• Stars: 64
• Watchers: 3
• Forks: 3
• Open Issues: 2
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Flow Models

The aim of this repository is to test and implement Flow-Matching-based models

# Warning

The data as well as dataset can not be shared!

## text2img generation example

Rectflow $^1$ full steps

![text2img](images/text2img.png)

Rectflow $^1$ 5 steps

![text2img](images/recflow_1_5_steps.png)

Rectflow $^1$ 1 step

![text2img](images/rectflow_1_1_step.png)

Rectflow $^2$ 5 steps

![text2img](images/recflow_2_5_steps.png)

Rectflow $^2$ 1 step

![text2img](images/rectflow_2_1_step.png)

Rectflow $^2$ + Distill 1 step

![text2img](images/rectflow_2_distill_1step.png)

Rectflow $^2$ + Distill + Midpoint 1 step

![text2img](images/rectflow_2_distill_1_step_midpoint.png)

LADD 4 steps

![text2img](images/rectflow_2_distill_4_steps_ladd.png)

Shortcut 4 steps

![text2img](images/shortcut.png)

## Inversion example

Input Image                                             |  FM Inversed                                                 |  Changed prompt

:------------------------------------------------------:|:------------------------------------------------------------:|:---------------------------------------:

 |   |  

# TODO:

- [x] [Train a foundint text2img model](sd_2_fm_finetuning.ipynb)

- [x] [FM Inversion](https://github.com/leffff/InstructFlow/blob/main/sd_2_fm_inversion.ipynb)

- [x] [Midpoint, Euler, RK4 solvers (with CFG)](https://github.com/leffff/InstructFlow/blob/main/instructflow/generation.py)

- [x] [N-th Reflow (for straightening and fast simulation)](https://github.com/leffff/InstructFlow/blob/main/sd_2_fm_finetuning_reflow_k.ipynb) (Same as Knowledge Distillaiton for Diffusion)

- [x] [Rectified Flow Distillation (1 step)](https://github.com/leffff/InstructFlow/blob/main/sd_2_fm_finetuning_reflow_distil_k.py)

- [x] [Immiscible Diffusion](https://github.com/leffff/FlowModels/blob/main/sd_2_fm_finetuning_immiscible.ipynb)

- [x] [Lattent Adversarial Diffusion Distillaiton](https://github.com/leffff/FlowModels/blob/main/sd_2_fm_finetuning_ladd.ipynb)

- [x] [Shortcut Model](https://github.com/leffff/FlowModels/blob/main/sd_2_fm_finetuning_shortcut.ipynb)

- [ ] Consisntency Distillation

- [ ] Phased (Multiboundary) Consisntency Distillation

- [ ] Diffusion RLHF

- [ ] Bridge Matching