Methods for processing a plurality of candidate annotations of a given instance of an image, and for learning parameters of a computational model
The present invention relates to a method for processing a plurality of candidate annotations of a given instance of an image, each candidate annotation being defined as a closed shape matching the instance, characterized in that the method comprises performing, by a processing unit ( 21 ) of a server ( 2 ), steps of: (a) segregating said candidate annotations into a set of separate groups of at least overlapping candidate annotations; (b) selecting a subset of said groups as a function of the number of candidate annotations in each group; (c) building a final annotation of the given instance of said image as a combination of regions of the candidate annotations of said selected groups where at least a second predetermined number of the candidate annotations of said selected groups overlap.
1. A method for processing a plurality of candidate annotations of a given instance of an image, each candidate annotation being defined as a closed shape matching the instance, wherein the method comprises performing, by a processing unit ( 21 ) of a server ( 2 ), steps of:
(a) segregating said candidate annotations into a set of separate groups, wherein each of the separate group consists of a single candidate annotation or consists of a plurality of overlapping candidate annotations;
(b) selecting a subset of said groups as a function of a first predetermined number of candidate annotations in each group;
(c) building a final annotation of the given instance of said image as a combination of regions of the candidate annotations of said selected groups, wherein at least a second predetermined number of the candidate annotations of said selected groups overlap in each of the regions.
2. A method according to claim 1 , wherein step (a) comprises successively, for each candidate annotation, either assigning it to an existing group of candidate annotations or creating a new group.
3. A method according to claim 1 , wherein two candidate annotations assigned to the same group present a Jaccard index above a predetermined threshold and/or are concentric.
4. A method according to claim 1 , wherein the groups selected at step (b) are the groups containing at least a first predetermined number of the candidate annotations.
5. A method according to claim 4 , wherein the first and the second predetermined number are respectively defined as a first rate of the number of all candidate annotations, and a second rate of the number of candidate annotations of said selected groups.
6. A method according to claim 1 , wherein the closed shape of each candidate annotation is a polygon.
7. A method according to claim 1 , wherein the final annotation of the given instance of said image is the union of all possible combinations of intersections of the second predetermined number of candidate annotations of said selected groups among all the candidate annotations of said selected groups.
8. A method according to claim 1 , wherein the candidate annotations are received from clients ( 1 ) of human annotators.
9. A method according to claim 8 , wherein said image is initially non-annotated, and the method comprises a previous step (a0) of transmitting the non-annotated image to the clients ( 1 ), and receiving in response from each client ( 1 ) a candidate annotation for at least one instance of the image, steps (a) to (c) being performed for each instance for which at least one candidate annotation has been received so as to build a final annotation for each instance annotated at least once of said image.
10. A method according to claim 9 , wherein said steps (a), (b), (c) are performed for a plurality of non-annotated images, so as to generate a training database for machine learning.
11. A method for learning parameters of a computational model comprising:
generating a training database using the method of claim 10 ;
performing a machine learning algorithm on said generated training database.
12. A computer program product stored on a non transitory computer-readable medium and comprising code instructions to execute a method according to claim 11 for learning parameters of a computational model, when said program is executed on a computer.
13. A non transitory computer-readable medium, on which is stored a computer program product comprising code instructions for executing a method according to claim 11 for learning parameters of a computational model.
14. A method according to claim 1 , comprising a step (d) of storing the image associated with the final annotation in a database.
15. A computer program product stored on a non transitory computer-readable medium and comprising code instructions to execute a method according to claim 1 for processing a plurality of candidate annotations of a given instance of an image.
16. A non transitory computer-readable medium, on which is stored a computer program product comprising code instructions for executing a method according to claim 1 for processing a plurality of candidate annotations of a given instance of an image.