System and method for generating training data sets for specimen defect detection

1. A method for generating a training data set for training a machine learning model to detect defects in specimens, the method comprising:

causing, by a computing system, presentation of a plurality of training images on a device of a user, wherein each training image of the plurality of training images comprises at least one training defect;

receiving, by the computing system, a plurality of training annotated images from the user based on the plurality of training images, wherein each plurality of training annotated images comprises a further indication of a location of each training defect;

learning, by the computing system, an individualized defect labeling pattern of the user based on the plurality of training annotated images;

generating, by the computing system, an error profile of the user based on the individualized defect labeling pattern of the user, wherein the error profile defines how to adjust annotations of the user;

causing, by the computing system, presentation of an image on the device of the user, the image comprising a defect on an example specimen;

receiving, by the computing system, an annotated image from the user, wherein the user annotated the image using an input via the device, wherein the input comprises a first indication of a location of the defect and a second indication of a class corresponding to the defect;

adjusting, by the computing system, the annotated image to standardize the input based on the error profile of the user and the class corresponding to the defect, wherein the error profile indicates how to adjust the first indication of the user; and

outputting, by the computing system, the annotated image for training the machine learning model.

2. The method of claim 1 , wherein the annotated image comprises a second input, the second input comprising a third indication of a second location of a second defect and a fourth indication of a second class corresponding to the second defect.

3. The method of claim 2 , further comprising:

filtering, by the computing system, the annotated image by class type.

4. The method of claim 1 , wherein adjusting, by the computing system, the annotated image to standardize the input based on the error profile of the user and the class corresponding to the defect comprises:

converting the first indication to a default shape; and

adjusting the default shape based on the error profile.

5. The method of claim 4 , further comprising:

transform the adjusted default shape into a second shape dictated by a type of the machine learning model.

6. The method of claim 1 , further comprising:

defining, by the computing system, a color pallet for the annotations, wherein each color in the color pallet corresponds to a specific class of defects.

7. A system, comprising:

a processor; and

a memory having programming instructions stored thereon, which, when executed by the processor, causes the processor to perform operations comprising:

causing presentation of a plurality of training images on a device of a user, wherein each training image of the plurality of training images comprises at least one training defect;

receiving a plurality of training annotated images from the user based on the plurality of training images, wherein each plurality of training annotated images comprises a further indication of a location of each training defect;

learning an individualized defect labeling pattern of the user based on the plurality of training annotated images;

generating an error profile of the user based on the individualized defect labeling pattern of the user, wherein the error profile defines how to adjust annotations of the user;

causing presentation of an image on the device of the user, the image comprising a defect on an example specimen;

receiving an annotated image from the user, wherein the user annotated the image using an input via the device, wherein the input comprises a first indication of a location of the defect and a second indication of a class corresponding to the defect;

adjusting the annotated image to standardize the input based on the error profile of the user and the class corresponding to the defect, wherein the error profile indicates how to adjust the first indication of the user; and

uploading the annotated image for training a machine learning model to detect defects on a specimen.

8. The system of claim 7 , wherein the annotated image comprises a second input, the second input comprising a third indication of a second location of a second defect and a fourth indication of a second class corresponding to the second defect.

9. The system of claim 8 , wherein the operations further comprise:

filtering the annotated image by class type.

10. The system of claim 7 , wherein adjusting the annotated image to standardize the input based on the error profile of the user and the class corresponding to the defect comprises:

converting the first indication to a default shape; and

adjusting the default shape based on the error profile.

11. The system of claim 10 , further comprising:

transform the adjusted default shape into a second shape dictated by a type of the machine learning model.

12. The system of claim 7 , wherein the operations further comprise:

defining a color pallet for the annotations, wherein each color in the color pallet corresponds to a specific class of defects.

13. A non-transitory computer readable medium comprising one or more sequences of instructions, which, when executed by a processor, causes a computing system to perform operations comprising:

causing, by the computing system, presentation of a plurality of training images on a device of a user, wherein each training image of the plurality of training images comprises at least one training defect;

receiving, by the computing system, a plurality of training annotated images from the user based on the plurality of training images, wherein each plurality of training annotated images comprises a further indication of a location of each training defect;

learning, by the computing system, an individualized defect labeling pattern of the user based on the plurality of training annotated images;

generating, by the computing system, an error profile of the user based on the individualized defect labeling pattern of the user, wherein the error profile defines how to adjust annotations of the user;

causing, by the computing system, presentation of an image on the device of the user, the image comprising a defect on an example specimen;

receiving, by the computing system, an annotated image from the user, wherein the user annotated the image using an input via the device, wherein the input comprises a first indication of a location of the defect and a second indication of a class corresponding to the defect;

adjusting, by the computing system, the annotated image to standardize the input based on the error profile of the user and the class corresponding to the defect, wherein the error profile indicates how to adjust the first indication of the user; and

uploading, by the computing system, the annotated image for training a machine learning model to detect defects on a specimen.

14. The non-transitory computer readable medium of claim 13 , wherein the annotated image comprises a second input, the second input comprising a third indication of a second location of a second defect and a fourth indication of a second class corresponding to the second defect.

15. The non-transitory computer readable medium of claim 14 , further comprising:

filtering, by the computing system, the annotated image by class type.

16. The non-transitory computer readable medium of claim 13 , wherein adjusting, by the computing system, the annotated image to standardize the input based on the error profile of the user and the class corresponding to the defect comprises:

converting the first indication to a default shape; and

adjusting the default shape based on the error profile.

17. The non-transitory computer readable medium of claim 16 , further comprising:

transform the adjusted default shape into a second shape dictated by a type of the machine learning model.