Trained machine learning model for estimating structure feature measurements
A computer system trains a machine learning model to estimate a real-world measurement of a feature of a structure. The machine learning model is trained using a plurality of digital image sets, wherein each image set depicts a particular structure, and a plurality of measurements, wherein each measurement is a measurement of a feature of a particular structure. After the machine learning model is trained, it is used to estimate a measurement of a feature of a particular structure depicted in a particular image set.
1. A method comprising:
training a machine learning model to estimate real-world measurements of features of real-world structures, wherein training the machine learning model includes providing to the machine learning model:
a plurality of image sets, wherein each image set of the plurality of image sets includes one or more 2-dimensional images of an exterior of a corresponding real-world structure, and
a plurality of real-world measurements, wherein the plurality of real-world measurements includes, for each image set of the plurality of image sets, a real-world measurement of a feature of the exterior of the corresponding real-world structure; and
after training the machine learning model using the machine learning model to estimate a particular real-world measurement of a particular feature of a particular real-world structure based on a particular digital image set containing one or more 2-dimensional images of an exterior of the particular real-world structure;
wherein the particular digital image set comprises metadata indicating a measurement of a second feature depicted in the one or more 2-dimensional images other than the particular feature, wherein the measurement of the second feature is a value based on an actual real-world distance of a dimension of the second feature;
wherein said using the machine learning model to estimate the particular real-world measurement of the particular feature comprises:
using the machine learning model to estimate the particular real-world measurement of the particular feature based, at least in part, on the measurement of the second feature, wherein the particular real-world measurement of the particular feature is a value based on an actual real-world distance of a dimension of the particular feature; and
estimating a roof pitch, wherein the particular digital image set for estimating the roof pitch includes one or more lateral photos of the exterior of the particular real-world structure and no orthographic photos of the exterior of the particular real-world structure.
2. The method of claim 1 wherein estimating the particular real-world measurement is based on a polynomial regression.
3. The method of claim 1 further comprising:
comparing the particular real-world structure with the plurality of real-world structures to identify how different the particular real-world structure is from the plurality of real-world structures;
based on how different the particular real-world structure is from the plurality of real-world structures, determining a confidence level associated with the particular real-world measurement.
4. The method of claim 1 wherein the particular digital image set includes an orthographic photo and one or more lateral photos.
5. The method of claim 1 wherein training the machine learning model is based on one or more metadata associated with at least one image in each image set of the plurality of image sets.
6. The method of claim 1 further comprising:
after training the machine learning model, using the machine learning model to estimate a second real-world measurement of a second feature of a second real-world structure based on a dataset comprising metadata describing second one or more images of the second real-world structure;
wherein using the machine learning model to estimate the second real-world measurement involves generating an estimate of the second real-world measurement without providing the machine learning model the second one or more images themselves.
7. The method of claim 1 wherein training the machine learning model is based on output from a second machine learning model.
8. The method of claim 1 wherein the particular real-world structure is a particular building and the plurality of image sets correspond to a plurality of buildings.
9. The method of claim 8 wherein the particular feature of the particular building is a roof of the particular building, and the particular real-world measurement is an area of the roof.
10. The method of claim 9 wherein:
the machine learning model is a first machine learning model; and
using the first machine learning model to estimate the particular real-world measurement comprises:
using the first machine learning model to estimate an outline of the roof;
using a second trained machine learning model to estimate a slope of the roof; and
calculating the area of the roof based on the outline of the roof and the slope of the roof.
11. The method of claim 1 , wherein the estimate of the particular real-world measurement of the particular feature is different than the measurement of the second feature.
12. The method of claim 1 , wherein the measurement of the second feature comprises output of a second machine learning model.
13. The method of claim 1 , wherein using the machine learning model to estimate the particular real-world measurement of the particular feature is further based, at least in part, on a spatial relationship between the particular feature and the second feature.
14. The method of claim 6 , wherein:
the dataset consists of the metadata describing the second one or more images of the second real-world structure; and
said using the machine learning model to estimate the second real-world measurement of the second feature of the second real-world structure is based exclusively on the dataset.
15. The method of claim 1 , wherein the particular real-world measurement of the particular feature is one or more of an actual real-world length of the particular feature, an actual real-world area of the particular feature, an actual real-world volume of the particular feature, and an actual real-world slope of the particular feature.
16. The method of claim 15 , wherein the length of the particular feature is a height or a perimeter length.
17. A system comprising:
one or more processors;
one or more non-transitory computer-readable media storing instructions which, when executed by the one or more processors, cause performance of:
training a machine learning model to estimate real-world measurements of features of real-world structures;
wherein training the machine learning model includes providing to the machine learning model:
a plurality of image sets, wherein each image set of the plurality of image sets includes one or more 2-dimensional images of an exterior of a corresponding real-world structure, and
a plurality of real-world measurements, wherein the plurality of real-world measurements includes, for each image set of the plurality of image sets, a real-world measurement of a feature of the exterior of the corresponding real-world structure; and
after training the machine learning model, using the machine learning model to estimate a particular real-world measurement of a particular feature of a particular real-world structure based on a particular digital image set containing one or more 2-dimensional images of an exterior of the particular real-world structure;
wherein the particular digital image set comprises metadata indicating a measurement of a second feature depicted in the one or more 2-dimensional images other than the particular feature, wherein the measurement of the second feature is a value based on an actual real-world distance of a dimension of the second feature;
wherein said using the machine learning model to estimate the particular real-world measurement of the particular feature comprises:
using the machine learning model to estimate the particular real-world measurement of the particular feature based, at least in part, on the measurement of the second feature, wherein the particular real-world measurement of the particular feature is a value based on an actual real-world distance of a dimension of the particular feature; and
estimating a roof pitch, wherein the particular digital image set for estimating the roof pitch includes one or more lateral photos of the exterior of the particular real-world structure and no orthographic photos of the exterior of the particular real-world structure.
18. The system of claim 17 wherein estimating the particular real-world measurement is based on a polynomial regression.
19. The system of claim 17 wherein the instructions, when executed, further cause:
comparing the particular real-world structure with the plurality of real-world structures to identify how different the particular real-world structure is from the plurality of real-world structures;
based on how different the particular real-world structure is from the plurality of real-world structures, determining a confidence level associated with the particular real-world measurement.
20. The system of claim 17 wherein the particular digital image set includes an orthographic photo and one or more lateral photos.
21. The system of claim 17 wherein training the machine learning model is based on one or more metadata associated with at least one image in each image set of the plurality of image sets.
22. The system of claim 17 wherein the instructions, when executed, further cause:
after training the machine learning model, using the machine learning model to estimate a second real-world measurement of a second feature of a second real-world structure based on a dataset comprising metadata describing second one or more images of the second real-world structure;
wherein using the machine learning model to estimate the second real-world measurement involves generating an estimate of the second real-world measurement without providing the machine learning model the second one or more images themselves.
23. The system of claim 17 wherein training the machine learning model is based on output from a second machine learning model.
24. The system of claim 17 wherein the particular real-world structure is a particular building and the plurality of image sets correspond to a plurality of buildings.
25. The system of claim 24 wherein the particular feature of the particular building is a roof of the particular building, and the particular real-world measurement is an area of the roof.
26. The system of claim 25 wherein:
the machine learning model is a first machine learning model; and
using the first machine learning model to estimate the particular real-world measurement comprises:
using the first machine learning model to estimate an outline of the roof;
using a second trained machine learning model to estimate a slope of the roof; and
calculating the area of the roof based on the outline of the roof and the slope of the roof.
27. The system of claim 17 , wherein the estimate of the particular real-world measurement of the particular feature is different than the measurement of the second feature.
28. The system of claim 17 , wherein the measurement of the second feature comprises output of a second machine learning model.
29. The system of claim 17 , wherein using the machine learning model to estimate the particular real-world measurement of the particular feature is further based, at least in part, on a spatial relationship between the particular feature and the second feature.
30. The system of claim 22 , wherein:
the dataset consists of the metadata describing the second one or more images of the second real-world structure; and
said using the machine learning model to estimate the second real-world measurement of the second feature of the second real-world structure is based exclusively on the dataset.
31. The system of claim 17 , wherein the particular real-world measurement of the particular feature is one or more of an actual real-world length of the particular feature, an actual real-world area of the particular feature, an actual real-world volume of the particular feature, and an actual real-world slope of the particular feature.
32. The system of claim 31 , wherein the length of the particular feature is a height or a perimeter length.