Orthophoto & Topographic Data: Bi-Modal Space Embeddings (2017-2022)

Open data collection for embeddings created upon research project called "City Knowledge Database" (POIR.01.01.01-00-1274/17-02):

grid_krakow_centroid.csvContains the definition of the grid for the city of Kraków, including cell identifiers (id) and boundary coordinates (left, top, right, bottom) in the EPSG:2180 coordinate system (PL-ETRS89).

grid_krakow_centroid_neighbors.csvContains the identifiers of neighboring cells for each cell in the Kraków grid. These data allow for the analysis of spatial relationships between cells within the city.

embeddings_krakow_2017-img_wizipisi_unetformer.pklContains image embeddings computed for individual grid cells using the Wizipisi_Unetformer network, based on the orthophoto map of Kraków from 2017. These embeddings represent the visual spatial features of the city's terrain.

embeddings_krakow_2017-img_txt_openai-clip-vit-base-path32.pklContains image embeddings computed for individual cells using the OpenAI CLIP-ViT-Base-Path32 network, based on the orthophoto map of Kraków from 2017. Additionally, it includes text embeddings for the same cells, computed based on textual descriptions of the terrain generated from the topographic map.

embeddings_krakow_2017-img_txt_n4_openai-clip-vit-base-path32.pklContains the average value of embeddings from the embeddings_krakow_2017-img_txt_openai-clip-vit-base-path32.pkl file, calculated by considering 4 neighboring cells for each cell in the grid.

embeddings_krakow_2017-img_txt_n8_openai-clip-vit-base-path32.pklContains the average value of embeddings from the embeddings_krakow_2017-img_txt_openai-clip-vit-base-path32.pkl file, calculated by considering 8 neighboring cells for each cell in the grid.

embeddings_krakow_2022-img_txt_openai-clip-vit-base-path32.pklContains image embeddings computed for individual cells using the OpenAI CLIP-ViT-Base-Path32 network, based on the orthophoto map of Kraków from 2022. Additionally, it includes text embeddings for the same cells, computed based on textual descriptions of the terrain generated from the topographic map.