User-generated visual guide for the classification of images
A process and a system for creating a visual guide for developing training data for a classification of image, where the training data includes images tagged with labels for the classification of the images. A processor may prompt a user to define a framework for the classification. For an initial set of images within the training data, qualified human classifiers are prompted to locate the images within the framework and to tag the images with labels. The processor determines whether the tagged images have consistent labels, and, if so, the processor adds images to the training data. The processor may add the images by providing a visual guide, the visual guide including tagged images arranged according to their locations within the framework their labels, and prompting human classifiers to tag the additional images with labels for the classification, according to the visual guide.
1. A computer-implemented method for developing training data for a classification of images, the training data comprising images tagged with labels for the classification of the images, the method comprising:
prompting a user to define a framework for the classification, the framework, when defined, including example images that are each paired with one or more example labels;
prompting qualified human classifiers to tag an initial set of images within the training data with labels for the classification based on the example labels within the framework; and
adding the tagged images to the training data, wherein the training data is used to train a machine learning model to automatically tag further images with labels according to the framework.
2. The computer-implemented method of claim 1 , further comprising generating a visual guide comprising tagged images arranged according the framework.
3. The computer-implemented method of claim 2 , the visual guide comprising a prompt to tag additional images with labels according to the framework, the additional images, when tagged, being added to the training data.
4. The method of claim 1 , wherein adding the tagged images to the training data is performed responsive to determining that the tagged images have consistent labels.
5. The computer-implemented method of claim 1 , further comprising:
determining whether an image in the initial set is weakly classifiable; and
responsive to determining that the image is weakly classifiable, modifying that image to be strongly classifiable.
6. The computer-implemented method of claim 1 , further comprising:
determining whether the label for a tagged image in the initial set is incorrect; and
responsive to determining that the label is incorrect, correcting the label for that image prior to adding that image to the training data.
7. The computer-implemented method of claim 1 , further comprising:
determining whether the tagged images provide sufficient samples across the framework; and
responsive to determining that the tagged images do not provide sufficient samples, prompting to provide additional tagged images in regions of the framework that have insufficient samples.
8. The computer-implemented method of claim 1 , wherein the framework comprises a spectrum of tagged images spanning a range of values for the labels, the method further comprising automatically placing a new image on the spectrum, based on a location of the new image within the framework and a similarity of the new image relative to the tagged images spanning the spectrum.
9. The computer-implemented method of claim 1 , further comprising using the training data to benchmark an accuracy of human classifiers, wherein the human classifiers are benchmarked against a machine learning model trained using the training data.
10. The computer-implemented method of claim 1 , wherein the framework dynamically adapts by giving more weight to newer images compared to older images.
11. A system comprising for developing training data for a classification of images, the training data comprising images tagged with labels for the classification of the images, the system comprising one or more processors that, when executing instructions, are caused perform operations comprising:
prompting a user to define a framework for the classification, the framework, when defined, including example images that are each paired with one or more example labels;
prompting qualified human classifiers to tag an initial set of images within the training data with labels for the classification based on the example labels within the framework; and
adding the tagged images to the training data, wherein the training data is used to train a machine learning model to automatically tag further images with labels according to the framework.
12. The system of claim 11 , the operations further comprising generating a visual guide comprising tagged images arranged according the framework.
13. The system of claim 12 , the visual guide comprising a prompt to tag additional images with labels according to the framework, the additional images, when tagged, being added to the training data.
14. The system of claim 11 , wherein adding the tagged images to the training data is performed responsive to determining that the tagged images have consistent labels.
15. The system of claim 11 , the operations further comprising:
determining whether an image in the initial set is weakly classifiable; and
responsive to determining that the image is weakly classifiable, modifying that image to be strongly classifiable.
16. The system of claim 11 , the operations further comprising:
determining whether the label for a tagged image in the initial set is incorrect; and
responsive to determining that the label is incorrect, correcting the label for that image prior to adding that image to the training data.
17. The system of claim 11 , the operations further comprising:
determining whether the tagged images provide sufficient samples across the framework; and
responsive to determining that the tagged images do not provide sufficient samples, prompting to provide additional tagged images in regions of the framework that have insufficient samples.
18. The system of claim 11 , wherein the framework comprises a spectrum of tagged images spanning a range of values for the labels, the operations further comprising automatically placing a new image on the spectrum, based on a location of the new image within the framework and a similarity of the new image relative to the tagged images spanning the spectrum.
19. The system of claim 11 , the operations further comprising using the training data to benchmark an accuracy of human classifiers, wherein the human classifiers are benchmarked against a machine learning model trained using the training data.
20. The system of claim 11 , wherein the framework dynamically adapts by giving more weight to newer images compared to older images.