earth-text VS SkyScript

What image size embeddings from what instrument?

We did try to relate OSM tags to Clay embeddings, but it didn't scale well. We did not give up, but we are re-considering ( https://github.com/Clay-foundation/earth-text ). I think SatClip plus OSM is a better approach. or LLM embeddings mapped to Clay embeddings...

Hey hey! We tried Clay v1 with 768 embeddings size using your tutorials. We then split NAIP SF to chips and indexed them. Afterwards, we performed image-to-image similarity search like in your explorer.

We tried to search for bridges, beaches, tennis courts, etc. It worked, but it didn't work well. The top of the ranking was filled with unrelated objects. We found that similarity scores are stacked together too much (similarity values are between 0.91 and 0.92 with 4 digit difference, ~200k tiles), so the encoder made very little difference between objects.

I believe that Clay can be used with additional fine-tuning for classification and segmentation, but standalone embeddings are pretty poor.

Check this: https://github.com/wangzhecheng/SkyScript. It is a dataset of OSM tags and satellite images. CLIP fine-tuned on that gives good embeddings for text-to-image search as well as image-to-image.

- Control the number of retrieved tiles with a slider

We use OpenAI's CLIP model (https://openai.com/index/clip/) to put texts and images into the same embedding space. We do a similarity search within this space using text query or source image. We are using CLIP finetuned on pairs of satellite images and OpenStreetMap (https://www.openstreetmap.org/) tags (https://github.com/wangzhecheng/SkyScript) because vanilla clip performs poorly on satellite data. We pre-segment objects using Meta's Segment Anything Model (https://segment-anything.com/) and pre-compute CLIP embeddings for each object.

We'd love to hear your thoughts! What worked well for you? Where did it fail? What features do you wish it had? Any real-world problems you think this could help with?