satclip
SatCLIP utilizes contrastive learning with satellite imagery for efficient geographic location encoding. Through a process similar to CLIP's image-text pairing, it accurately trains both location and image encoders. SatCLIP supports the S2-100K dataset, featuring cloud-free, multi-spectral Sentinel-2 images, and offers six pretrained models fit for different resolutions and tasks. Its application spans air temperature prediction and image localization. Integration with Hugging Face facilitates easy deployment and experimentation.
