GeoCR: Learning a Generalist Cloud Removal Prior from Heterogeneous Observations
GeoCR is a generalist cloud-removal model unifying RGB and multispectral inputs, with optional SAR, across ten datasets.
GeoCR unifies RGB-only and multispectral cloud removal from single- or multi-temporal cloudy observations, with optional SAR guidance, in one network. Compact stems extend a frozen pretrained RGB autoencoder so a flow transformer jointly models clean RGB and non-RGB latents. Joint pretraining used ten datasets totaling 883,331 cloud-free target images. The checkpoint supports inference without dataset-specific fine-tuning and LoRA adaptation, with the best reported FID and DISTS on full-band SEN12MS-CR and Sen2_MTC_New and RGB-only CUHK-CR2.
- Unifies RGB, multispectral, and optional SAR cloud removal in one network.
- Frozen RGB autoencoder trunks plus compact stems and a flow transformer.
- Pretrained on 883,331 cloud-free targets from ten datasets.
- Best FID and DISTS on SEN12MS-CR, Sen2_MTC_New, and CUHK-CR2.
- Same checkpoint supports direct inference and LoRA adaptation.
From the original article
1 min read at huggingface.co
Cloud removal methods are typically specialized to individual datasets and input configurations, limiting reuse across sensors, spectral bands, and observation settings. We introduce GeoCR, a generalist model that unifies RGB-only-based CR and multispectral-based CR from single- or multi-temporal cloudy observations, with optional SAR guidance, within a single network.
By Jeonghyeok Do, Munchurl Kim · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.