BETACLARITY β BetaSR v1
Medical image super-resolution and denoising powered by latent diffusion.
π₯ New: Edge-optimized version available! A quantized ONNX bundle (UNet INT8, 3.97x smaller, no PyTorch required at runtime) is now published at
OdaxAI/betaclarity-betasr-onnx. It runs natively on Apple Silicon via the CoreML Execution Provider. Quantized with our open-source OdaxAI SDK.
Research use only. This model is not intended for clinical decision-making. See the Disclaimer below.
Model description
BetaSR is the core model of BETACLARITY β a latent diffusion pipeline fine-tuned on synthetic degradations (noise + downscaling) applied to multi-modality medical images.
It wraps CompVis/ldm-super-resolution-4x-openimages and fine-tunes the UNet using:
- Wavelet-domain regularization β selective masking of high-frequency wavelet coefficients
- Dual masking strategy β latent random masking + wavelet coefficient masking
- DDIM controllable inference β 5β50 steps for quality/speed trade-off
Supported modalities: chest X-ray, brain MRI, cardiac MRI, breast MRI, knee MRI, CT, ultrasound.
Downloads
huggingface-cli download OdaxAI/betaclarity-betasr model.pth --local-dir ./weights
from huggingface_hub import hf_hub_download
model_path = hf_hub_download(
repo_id="OdaxAI/betaclarity-betasr",
filename="model.pth",
local_dir="./weights",
)
Usage
from betaclarity.core.models import EnhancedLatentDiffusionModel
from diffusers import LDMSuperResolutionPipeline
import torch
pipeline = LDMSuperResolutionPipeline.from_pretrained(
"CompVis/ldm-super-resolution-4x-openimages"
)
model = EnhancedLatentDiffusionModel(pipeline=pipeline)
checkpoint = torch.load("./weights/model.pth", map_location="cpu")
model.unet.load_state_dict(checkpoint["model_state_dict"])
model.eval()
enhanced = model.reconstruct(degraded_tensor, ddim_steps=10)
Full documentation and training instructions: github.com/odaxai/BETACLARITY
Model card
| Property | Value |
|---|---|
| Base model | CompVis/ldm-super-resolution-4x-openimages |
| Task | Medical image super-resolution and denoising |
| Scale factor | 4Γ |
| Input | 128Γ128 low-quality image |
| Output | 512Γ512 enhanced image |
| Framework | PyTorch + Hugging Face Diffusers |
| License | Apache License 2.0 |
Disclaimer
The model described in this repository is provided for research and development use only.
The model is not intended for use in clinical decision-making or for any other clinical use, and the performance of the model for clinical use has not been established. You bear sole responsibility for any use of this model, including incorporation into any product intended for clinical use.
This software does not constitute a medical device. It has not been cleared or approved by any regulatory agency (FDA, EMA, etc.).
License
Apache License 2.0 β see LICENSE.
Copyright 2026 OdaxAI SRL. All rights reserved.