Partial labeling mechanism for quick and accurate training of machine learning models
A device generates a training set of images by, for each image of a plurality of training images, receiving user input of a set of labels for a portion of the image, the portion less than an entirety of the image, the set of labels comprising classifications of individual pixels within the image, and automatically applying a label of unknown to a remainder of the image that excludes the portion of the image. The device inputs an unlabeled image into a machine learning model, the machine learning model trained using the training set, and receives, as output from the machine learning model, predicted classifications for each pixel of the image.
1 . A method comprising:
generating a training set of images by, for each image of a plurality of training images:
receiving user input of a set of labels for a portion of the image, the portion less than an entirety of the image, the set of labels comprising classifications of individual pixels within the image; and
automatically applying a label of unknown to a remainder of the image that excludes the portion of the image;
inputting an unlabeled image into a machine learning model, the machine learning model trained using the training set, wherein a manner in which the machine learning model is trained using the training set comprises:
generating a plurality of feature vectors comprising a feature vector for each pixel;
mapping each feature vector to a label corresponding to the pixel from which each feature vector was derived; and
using the plurality of feature vectors to train the machine learning model, wherein using the plurality of feature vectors to train the machine learning model comprises, for each different label of the training set:
generating a core set of feature vectors, the core set comprising fewer than all of the feature vectors mapped to the label; and
using the core set to train the machine learning model, wherein feature vectors for the label that are not within the core set are not used to train the machine learning model; and
receiving, as output from the machine learning model, predicted classifications for each pixel of the image.
2 . The method of claim 1 , wherein receiving the user input of the set of labels comprises receiving, by way of a graphical user interface having a display of the image, input from the user that classifies the individual pixels of the image.
3 . The method of claim 2 , wherein the input simultaneously indicates a same classification for two or more of the individual pixels of the image.
4 . The method of claim 1 , wherein generating the core set of feature vectors for a given label comprises:
obtaining a subset of the plurality of feature vectors by sampling the plurality of feature vectors; and
adding additional ones of the plurality of feature vectors to the subset until a criterion is reached.
5 . The method of claim 4 , wherein adding additional ones of the plurality of feature vectors to the subset until the criterion is reached comprises:
identifying, for a given feature vector of the subset, a furthest feature vector of the plurality of feature vectors that is not already in the subset; and
adding the furthest feature vector to the subset.
6 . The method of claim 5 , wherein the criterion is a predefined amount of feature vectors required to form the core set.
7 . The method of claim 1 , wherein using the core set to train the machine learning model comprises applying a semi-supervised training technique to the core set.
8 . The method of claim 1 , further comprising:
receiving, from a user, further input indicating that a classification is incorrect;
in response to receiving the further input, re-labeling images of the training set having pixels with the classification with an adjusted classification and re-training the machine learning model.
9 . A non-transitory computer-readable medium comprising memory with instructions encoded thereon that, when executed by one or more processors, cause the one or more processors to perform operations, the instructions comprising instructions to:
generate a training set of images by, for each image of a plurality of training images:
receiving user input of a set of labels for a portion of the image, the portion less than an entirety of the image, the set of labels comprising classifications of individual pixels within the image; and
automatically applying a label of unknown to a remainder of the image that excludes the portion of the image;
input an unlabeled image into a machine learning model, the machine learning model trained using the training set, wherein a manner in which the machine learning model is trained using the training set comprises:
generating a plurality of feature vectors comprising a feature vector for each pixel;
mapping each feature vector to a label corresponding to the pixel from which each feature vector was derived; and
using the plurality of feature vectors to train the machine learning model, wherein using the plurality of feature vectors to train the machine learning model comprises, for each different label of the training set:
generating a core set of feature vectors, the core set comprising fewer than all of the feature vectors mapped to the label; and
using the core set to train the machine learning model, wherein feature vectors for the label that are not within the core set are not used to train the machine learning model; and
receive, as output from the machine learning model, predicted classifications for each pixel of the image.
10 . The non-transitory computer-readable medium of claim 9 , wherein the instructions to receive the user input of the set of labels comprise instructions to receive, by way of a graphical user interface having a display of the image, input from the user that classifies the individual pixels of the image.
11 . The non-transitory computer-readable medium of claim 10 , wherein the input simultaneously indicates a same classification for two or more of the individual pixels of the image.
12 . The non-transitory computer-readable medium of claim 9 , wherein generating the core set of feature vectors for a given label comprises:
obtaining a subset of the plurality of feature vectors by sampling the plurality of feature vectors; and
adding additional ones of the plurality of feature vectors to the subset until a criterion is reached.
13 . The non-transitory computer-readable medium of claim 12 , wherein adding additional ones of the plurality of feature vectors to the subset until the criterion is reached comprises:
identifying, for a given feature vector of the subset, a furthest feature vector of the plurality of feature vectors that is not already in the subset; and
adding the furthest feature vector to the subset.
14 . The non-transitory computer-readable medium of claim 13 , wherein the criterion is a predefined amount of feature vectors required to form the core set.
15 . The non-transitory computer-readable medium of claim 9 , wherein using the core set to train the machine learning model comprises applying a semi-supervised training technique to the core set.
16 . A system comprising:
memory with instructions encoded thereon; and
one or more processors that, when executing the instructions, are caused to perform operations comprising:
generating a training set of images by, for each image of a plurality of training images:
receiving user input of a set of labels for a portion of the image, the portion less than an entirety of the image, the set of labels comprising classifications of individual pixels within the image; and
automatically applying a label of unknown to a remainder of the image that excludes the portion of the image;
inputting an unlabeled image into a machine learning model, the machine learning model trained using the training set, wherein a manner in which the machine learning model is trained using the training set comprises:
generating a plurality of feature vectors comprising a feature vector for each pixel;
mapping each feature vector to a label corresponding to the pixel from which each feature vector was derived; and
using the plurality of feature vectors to train the machine learning model, wherein using the plurality of feature vectors to train the machine learning model comprises, for each different label of the training set:
generating a core set of feature vectors, the core set comprising fewer than all of the feature vectors mapped to the label; and
using the core set to train the machine learning model, wherein feature vectors for the label that are not within the core set are not used to train the machine learning model; and
receiving, as output from the machine learning model, predicted classifications for each pixel of the image.