1
AI use in Regional Simulation (BRAILS) / Questions for window/door detection
« on: January 30, 2023, 08:26:14 PM »
Hello,
I'm trying to get street-view building images using BRAILS and detect windows and doors from collected building images using FacadeParser.
Q1. It seems BRAILS finds building locations from the database of OpenStreetMap. If my buildings are not registered in the database of OpenStreetMap, is it impossible to get street-view images automatically using Google API?
Q2. As you can see in the attached image file, the accuracy of detecting windows and doors is great. However, there are still errors, for example, wrong identification of a door (the small door next to the actual door). Also, the detected center biggest window seem much smaller than its actual size. Is this common level of errors we can encounter while using FacadeParser? And, is there any other module or method to detect windows and doors?
Q3. There are three buildings in the attached image whereas I only need the segmented result of the center building. Do you have any idea to define the boundary of the center building and discard the images of other two buildings?
I'm trying to get street-view building images using BRAILS and detect windows and doors from collected building images using FacadeParser.
Q1. It seems BRAILS finds building locations from the database of OpenStreetMap. If my buildings are not registered in the database of OpenStreetMap, is it impossible to get street-view images automatically using Google API?
Q2. As you can see in the attached image file, the accuracy of detecting windows and doors is great. However, there are still errors, for example, wrong identification of a door (the small door next to the actual door). Also, the detected center biggest window seem much smaller than its actual size. Is this common level of errors we can encounter while using FacadeParser? And, is there any other module or method to detect windows and doors?
Q3. There are three buildings in the attached image whereas I only need the segmented result of the center building. Do you have any idea to define the boundary of the center building and discard the images of other two buildings?