---
library_name: none
license: mit
pipeline_tag: image-to-image
tags:
- image-super-resolution
- remote-sensing
- satellite-imagery
- cloud-microphysics
---

# Recovering Cloud Microstructures with Cascaded Diffusion Inversion

This repository contains the model checkpoints for **CloudSR**, a two-stage diffusion-based super-resolution framework to enhance the resolution of multi-spectral cloud microstructures by a factor of 4×. 

- **Paper:** [Recovering Cloud Microstructures with Cascaded Diffusion Inversion](https://huggingface.co/papers/2607.05637)
- **Project Page:** [Project Website](https://hananshafi.github.io/superresolution-cloud-microphysics/)
- **Code:** [GitHub Repository](https://github.com/hananshafi/superresolution-cloud-microphysics)

## Checkpoint Files

- `cloudsr_seviri_to_viirs_model_50000.pth`
  Single-stage checkpoint for SEVIRI to VIIRS super-resolution.

- `cloudsr_msg_to_mtg_model_50000.pth`
  Single-stage checkpoint for MSG to MTG super-resolution.

## Usage

These are custom PyTorch checkpoints intended to be used with the local inference code in the main project repository.

### Setup

First, clone the repository and install the dependencies:

```bash
git clone https://github.com/hananshafi/superresolution-cloud-microphysics.git
cd superresolution-cloud-microphysics
conda env create -f environment.yaml
conda activate cloudsr
```

### Download Checkpoints

You can download the checkpoints directly using `huggingface_hub`:

```bash
pip install huggingface_hub
hf download hanangani/cloudsr-checkpoints cloudsr_seviri_to_viirs_model_50000.pth --repo-type model --local-dir ./checkpoints
hf download hanangani/cloudsr-checkpoints cloudsr_msg_to_mtg_model_50000.pth --repo-type model --local-dir ./checkpoints
```

### SEVIRI to VIIRS Inference

Run inference using the `inference_sr.py` script:

```bash
python inference_sr.py \
  -i /path/to/seviri_input \
  -o /path/to/output_dir \
  --num_steps 1 \
  --sd_path /path/to/sd-turbo \
  --started_ckpt_path ./checkpoints/cloudsr_seviri_to_viirs_model_50000.pth
```

### MSG to MTG Inference

Run inference using the `inference_msg_to_mtg_sr.py` script:

```bash
python inference_msg_to_mtg_sr.py \
  -i /path/to/msg_input \
  -o /path/to/output_dir \
  --num_steps 1 \
  --sd_path /path/to/sd-turbo \
  --started_ckpt_path ./checkpoints/cloudsr_msg_to_mtg_model_50000.pth
```

## Citation

```bibtex
@inproceedings{gani2026recovering,
  title     = {Recovering Cloud Microstructures with Cascaded Diffusion Inversion},
  author    = {Gani, Hanan and Pulik, Guy and Rosenfeld, Daniel and Watson-Parris, Duncan and Khan, Salman},
  booktitle = {ICLR 2026 Workshop on Machine Learning for Remote Sensing (ML4RS)},
  year      = {2026}
}
```