Echo-DM / README.md
FelixCoooper's picture
Prepare Echo-DM gated research release
1a7fcc7 verified
|
Raw History Blame Contribute Delete
4.45 kB
---
license: other
license_name: echo-dm-non-commercial-research-access
license_link: https://huggingface.co/FelixCoooper/Echo-DM/blob/main/USE_CONDITIONS.md
pipeline_tag: image-to-image
tags:
- ultrasound
- image-restoration
- diffusion
- marker-removal
extra_gated_heading: "Request access for non-commercial research"
extra_gated_description: "Requests are reviewed manually. Provide your institutional affiliation, an institutional email address you control, and a specific research purpose."
extra_gated_button_content: "Submit access request"
extra_gated_prompt: "By submitting an application, you acknowledge the use conditions and agree that your application and shared Hugging Face account information may be used to assess and administer your access request."
extra_gated_fields:
"Full name": text
"Institution / organization": text
"Department / laboratory": text
"Institutional email address": text
"Institutional profile / laboratory URL": text
"Role":
type: select
options:
- Student
- Researcher
- Faculty
- Clinician
- Other
"Research project and intended non-commercial use": text
"I confirm that I control the institutional email address provided": checkbox
"I have read the use conditions and agree to non-commercial research use only": checkbox
---
# Echo-DM-V (Efficient)
Checkpoint bundle for **Echo-DM: Ultrasound Marker Removal via Conditional Latent
Diffusion and Region-Aware Fusion**. Only the Efficient configuration is included.
| Setting | Value |
|---|---|
| DiT input size | 512 x 512 |
| VAE compression | 8x |
| Sampling steps | 50 |
| Latent multiplier | 0.932 |
| RAF | 2x-resolution fusion with residual refinement |
## Files
```text
vae/8x/config.json
vae/8x/diffusion_pytorch_model.safetensors
v-efficient/dit.pt
v-efficient/raf.pt
manifest.json
```
The four checkpoint/config files total approximately 859 MB. The manifest records
their byte sizes and SHA-256 checksums. This repository supplies component weights;
run the pipeline using the GitHub code rather than loading this repository as a
standalone Diffusers pipeline.
## Download and run after approval
From the inference-code repository root, with `huggingface_hub` installed:
```bash
hf auth login
hf download FelixCoooper/Echo-DM --repo-type model --local-dir weights
python -m echo_dm.inference --input /path/to/marked.png --output outputs/example --seed 42
```
Inference requires only marked images, and output PNGs retain their original dimensions.
See the GitHub README for installation and usage.
## Use conditions and upstream components
Non-commercial research access is subject to [USE_CONDITIONS.md](USE_CONDITIONS.md).
The Apache 2.0 license for original code contributions in the GitHub repository
does not license this checkpoint bundle. Upstream component terms remain
applicable; this bundle uses an adapted [SD VAE](https://huggingface.co/stabilityai/sd-vae-ft-ema).
## Request access
1. Sign in to your individual Hugging Face account and submit the application.
2. Supply your institution, institutional email, official profile and a specific
non-commercial research purpose. Accept the [use conditions](USE_CONDITIONS.md).
3. Wait for manual review. Verification of institutional affiliation and control
of the institutional mailbox is required before approval. Additional verification
instructions may be provided privately during review.
4. Once approved, authenticate downloads with the approved account. Approval is
per individual user and per repository; model and dataset permissions are separate.
The institutional-email form field is self-reported; submitting the form does
not itself verify ownership of that mailbox. Do not post application details
in public GitHub Issues or share your account tokens.
## Application information
The publisher uses the submitted information and shared HF username/account email
for access review and administration. Application records should be accessible
only to authorized reviewers. Additional privacy/retention details will be
provided before applications are opened publicly.
## Related resources
- [Paper](https://arxiv.org/abs/2606.09378)
- [Publisher article](https://www.sciencedirect.com/science/article/abs/pii/S0893608026012128)
- [Inference code and documentation](https://github.com/MiliLab/Echo-DM)
- [Evaluation data](https://huggingface.co/datasets/FelixCoooper/Echo-PAIR)