Spaces:
Running on Zero
Running on Zero
|
Download ATTRIBUTION.md from SammySnake/CLIP: direct link, hf CLI and curl.
- Browser
- Download file 2.11 kB
-
https://huggingface.co/spaces/SammySnake/CLIP/resolve/main/ATTRIBUTION.md
- Command line
-
hf download hf://spaces/SammySnake/CLIP/ATTRIBUTION.md
-
curl -L -o ATTRIBUTION.md https://huggingface.co/spaces/SammySnake/CLIP/resolve/main/ATTRIBUTION.md
2.11 kB
| # Attribution and usage limits | |
| ## StreetCLIP | |
| - Authors: Lukas Haas, Silas Alberti, Michal Skreta. | |
| - Work: *Learning Generalized Zero-Shot Learners for Open-Domain Image Geolocalization* (2023). | |
| - Model: https://huggingface.co/geolocal/StreetCLIP | |
| - Model revision: `e3561ba2ad9bf14c9efd6b0092607b8497efbfea`. | |
| - License: [Creative Commons Attribution-NonCommercial 4.0](https://creativecommons.org/licenses/by-nc/4.0/). | |
| - This app uses the model as published; no weight changes. Added bounded CPU scoring, JSON contracts and UI are application code, not original model features. | |
| - Training set excludes China and India and is biased toward western countries. The model card discourages geolocating users' private images. Use only authorized scene photos and do not use the app to track people. | |
| ## GeoCLIP | |
| - Authors: Vicente Vivanco Cepeda, Gaurav Kumar Nayak, Mubarak Shah. | |
| - Work: *GeoCLIP: CLIP-Inspired Alignment between Locations and Images for Effective Worldwide Geo-localization*, NeurIPS 2023. | |
| - Source: https://github.com/VicenteVivan/geo-clip | |
| - Package: `geoclip==1.2.3`; inspected source commit `7c102b4647f031a50be0c85a2b3cd4f96ddc3add`. | |
| - Published trained head / GPS encoder and coordinate gallery: https://huggingface.co/VicenteVivanco/geoclip-geolocation | |
| - Model revision: `40a931123d2c6451ef7da1d37cf68cded0ec4d35`. | |
| - License: MIT; original model license retained in `LICENSE-GeoCLIP.txt`. | |
| - Original weights are unchanged. The app precomputes normalized gallery location embeddings in batches and applies the same normalized dot product / learned logit-scale formula as upstream `GeoCLIP.forward`. | |
| - Backbone: OpenAI CLIP ViT-L/14, pinned by the author's release config to `32bd64288804d66eefd0ccbe215aa642df71cc41`. | |
| ## Source and score boundaries | |
| GeoCLIP aligns photos with geographic coordinate embeddings. StreetCLIP aligns photos with text descriptions. Neither is a ground-to-satellite cross-view retrieval checkpoint or an RGB bird's-eye photograph reconstruction model. No accuracy claim, stable free hosting guarantee, or commercial usage permission is implied by this deployment. | |