Face detection to address privacy in publishing image datasets

Abstract

Methods for face detection to address privacy in publishing image datasets is described. A method may include face classification in an online marketplace. A server system may receive, from a seller user device, a listing including an image for the online marketplace. The server system may classify, by at least one processor that implement a distribution-balance trained machine learning model, each human face candidate within the image as being one of a private human face or a non-private human face. The server system may receive, from a buyer user device, a search query that is mapped to the listing in the online marketplace. The server system may transmit, to the buyer user device, a query response including the listing that includes the image determined to not include any private human faces or obscures any private human faces within the image based on the classifying.

What is claimed is: 
     
         1 . A system for face classification, comprising:
 one or more processors; and   a computer readable medium storing instructions that, when executed by the one or more processors, cause the system to perform operations comprising:
 receiving at least one image; 
 classifying, by a trained machine learning model, each human face candidate within the at least one image as being one of a private human face or a non-private human face; and 
 outputting, to a client device, a first image of the at least one image determined to not include any private human faces or that obscures any private human faces within the first image based at least in part on the classifying. 
   
     
     
         2 . The system of  claim 1 , the operations further comprising:
 receiving a search query that corresponds to the first image; and   transmitting, based at least in part on the search query, a query response comprising the first image based at least in part on each human face candidate within the first image being classified as a non-private human face.   
     
     
         3 . The system of  claim 1 , the operations further comprising:
 receiving a search query that corresponds to the first image; and   transmitting, based at least in part on the search query, a query response comprising the first image based at least in part on each human face classified as a non-private human face being obscured within the first image.   
     
     
         4 . The system of  claim 1 , the operations further comprising:
 receiving a search query that corresponds to a second image; and   transmitting, based at least in part on the search query, a query response without the second image based at least in part on at least one human face candidate within the second image being classified as a private human face.   
     
     
         5 . The system of  claim 1 , the operations further comprising:
 generating the trained machine learning model using a plurality of training images that are associated with a plurality of training scores indicating whether individual training images of the plurality of training images include at least one private human face or do not include any private human faces.   
     
     
         6 . The system of  claim 1 , the operations further comprising:
 generating the trained machine learning model using a distribution-balance value determined for a plurality of training images.   
     
     
         7 . The system of  claim 6 , wherein the distribution-balance value is a weight value applied to a difference between one or more training scores of a plurality of training scores and one or more predicted scores of a plurality of predicted scores. 
     
     
         8 . The system of  claim 1 , wherein the non-private human face comprises a non-human face, a published human face, a public figure human face, a drawing of a human face, or any combination thereof. 
     
     
         9 . A computer-implemented method for face classification, comprising:
 receiving at least one image;   classifying, by at least one processor that implements a trained machine learning model, each human face candidate within the at least one image as being one of a private human face or a non-private human face; and   outputting, to a client device, a first image of the at least one image determined to not include any private human faces or obscures any private human faces within the first image based at least in part on the classifying.   
     
     
         10 . The computer-implemented method of  claim 9 , further comprising:
 receiving a search query that corresponds to the first image; and   transmitting, based at least in part on the search query, a query response comprising the first image based at least in part on each human face candidate within the first image being classified as a non-private human face.   
     
     
         11 . The computer-implemented method of  claim 9 , further comprising:
 receiving a search query that corresponds to the first image; and   transmitting, based at least in part on the search query, a query response comprising the first image based at least in part on each human face classified as a non-private human face being obscured within the first image.   
     
     
         12 . The computer-implemented method of  claim 9 , further comprising:
 receiving a search query that corresponds to a second image; and   transmitting, based at least in part on the search query, a query response without the second image based at least in part on at least one human face candidate within the second image being classified as a private human face.   
     
     
         13 . The computer-implemented method of  claim 9 , further comprising:
 generating the trained machine learning model using a plurality of training images that are associated with a plurality of training scores indicating whether individual training images of the plurality of training images include at least one private human face or do not include any private human faces.   
     
     
         14 . The computer-implemented method of  claim 9 , further comprising:
 generating the trained machine learning model using a distribution-balance value determined for a plurality of training images.   
     
     
         15 . The computer-implemented method of  claim 14 , wherein the distribution-balance value is a weight value applied to a difference between one or more training scores of a plurality of training scores and one or more predicted scores of a plurality of predicted scores. 
     
     
         16 . The computer-implemented method of  claim 9 , wherein the non-private human face comprises a non-human face, a published human face, a public FIG. human face, a drawing of a human face, or any combination thereof. 
     
     
         17 . A non-transitory computer-readable medium storing code for face classification, the code comprising instructions, when executed by a processor, cause a system to perform operations comprising:
 receiving at least one image;   classifying, by a trained machine learning model, each human face candidate within the at least one image as being one of a private human face or a non-private human face; and   outputting, to a client device, a first image of the at least one image determine to not include any private human faces or obscures any private human faces within the first image based at least in part on the classifying.   
     
     
         18 . The non-transitory computer-readable medium of  claim 17 , the operations further comprising:
 receiving a search query that corresponds to the first image; and   transmitting, based at least in part on the search query, a query response comprising the first image based at least in part on each human face candidate within the first image being classified as a non-private human face.   
     
     
         19 . The non-transitory computer-readable medium of  claim 17 , the operations further comprising:
 receiving a search query that corresponds to the first image; and   transmitting, based at least in part on the search query, a query response comprising the first image based at least in part on each human face classified as a non-private human face being obscured within the first image.   
     
     
         20 . The non-transitory computer-readable medium of  claim 17 , the operations further comprising:
 receiving a search query that corresponds to a second image; and   transmitting, based at least in part on the search query, a query response without the second image based at least in part on at least one human face candidate within the second image being classified as a private human face.