Object localization framework for unannotated image data
A system is provided for object localization in image data. The system includes an object localization framework comprising a plurality of object localization processes. The system is configured to receive an image comprising unannotated image data having at least one object in the image, access a first object localization process of the plurality of object localization processes, determine first bounding box information for the image using the first object localization process, wherein the first bounding box information comprises at least one first bounding box annotating at least a first portion of the at least one object in the image, and receive first feedback regarding the first bounding box information determined by the first object localization process. The system is further configured to persist the image with the first bounding box information or access a second object localization process based on the first feedback.
1. A system for identifying objects in image data, the system comprising:
an input interface configured to receive the image data including an object in one or more images corresponding to the image data;
a non-transitory memory storing a neural network based object localization model and a plurality of processor-executable instructions for generating annotation data including bounding boxes for the object using one or more bounding box operations; and
one or more hardware processors execute the plurality of processor-executable instructions to perform operations comprising:
generating a first set of annotation data including a first bounding box for the object using a first bounding box operation
receiving, via the input interface, feedback on the first set of annotation data; and
determining whether to generate a second set of annotation data including a second bounding box for the object using a second bounding box operation based on the feedback on the first set of annotation data.
2. The system of claim 1 , wherein the first bounding box operation comprises utilizing the neural network to generate convolution layer heatmaps of the object in the one or more images.
3. The system of claim 1 , wherein the in response to the feedback being the first bounding box not representing the object in the one or more images:
generating the second bounding box for the object using the second bounding box operation, wherein the second bounding box operation comprises using basic contours in the one or more images from Haar cascades;
receiving, via the input interface, second feedback on the second set of annotation data; and
determining whether to generate a third set of annotation data including a third bounding box for the object using a third bounding box operation based the second feedback on the second set of annotation data.
4. The system of claim 3 , wherein in response to the second feedback being the second bounding box not representing the object in the one or more images:
generating the third bounding box for the object using the third bounding box operation, wherein the third bounding box operation comprises converting the one or more images using canny edge detection and applying basic contours in the one or more converted images from Haar cascades;
receiving, via the input interface, third feedback on the third set of annotation data; and
determining whether to generate a fourth set of annotation data including a fourth bounding box for the object using a fourth bounding box operation based the third feedback on the third set of annotation data.
5. The system of claim 4 , wherein in response to the third feedback being the third bounding box not representing the object in the one or more images:
generating the fourth bounding box for the object using the fourth bounding box operation, wherein the fourth bounding box operation comprises applying image erosion to the one or more images and highlighting margins among the at least one object in the one or more images after the image erosion;
receiving, via the input interface, fourth feedback on the fourth set of annotation data; and
determining whether to output the fourth set of annotation data including the fourth bounding box based on the fourth feedback.
6. The system of claim 1 , wherein in response to the feedback being the first bounding box representing the object in the one or more images, outputting the first set of annotation data.
7. The system of claim 1 , wherein the neural network based object localization framework is trained using an unsupervised learning operation with unannotated image data for a plurality of objects.
8. A method for identifying objects in image data, the method comprising:
receiving the image data including an object in one or more images corresponding to the image data;
generating, via a neural network based object localized framework, a first set of annotation data including a first bounding box for the object using a first bounding box operation;
receiving feedback on the first set of annotation data; and
determining whether to generate, via the neural network based object localized framework, a second set of annotation data including a second bounding box for the object using a second bounding box operation based on the feedback on the first set of annotation data.
9. The method of claim 8 , wherein the first bounding box operation comprises utilizing the neural network to generate convolution layer heatmaps of the object in the one or more images.
10. The method of claim 8 , wherein in response to the feedback being the first bounding box not representing the object in the one or more images:
generating the second bounding box for the object using the second bounding box operation, wherein the second bounding box operation comprises using basic contours in the one or more images from Haar cascades;
receiving second feedback on the second set of annotation data; and
determining whether to generate, via the neural network based object localized framework, a third set of annotation data including a third bounding box for the object using a third bounding box operation based on the second feedback on the second set of annotation data.
11. The method of claim 10 , wherein in response to the second feedback being the second bounding box not representing the object in the one or more images:
generating the third bounding box for the object using the third bounding box operation, wherein the third bounding box operation comprises converting the one or more images using canny edge detection and applying basic contours in the one or more converted images from Haar cascades;
receiving third feedback on the third set of annotation data; and
determining whether to generate, via the neural network based object localized framework, a fourth set of annotation data including a fourth bounding box for the object using a fourth bounding box operation based on the third feedback on the third set of annotation data.
12. The method of claim 11 , wherein in response to the third feedback being the third bounding box not representing the object in the one or more images:
generating the fourth bounding box for the object using the fourth bounding box operation, wherein the fourth bounding box operation comprises applying image erosion to the one or more images and highlighting margins among the at least one object in the one or more images after the image erosion;
receiving fourth feedback on the fourth set of annotation data; and
determining whether to output the fourth set of annotation data including the fourth bounding box based on the fourth feedback.
13. The method of claim 8 , wherein in response to the first bounding box representing the object in the one or more images, outputting the first set of annotation data.
14. The method of claim 8 , wherein the neural network based object localization framework is trained using an unsupervised learning operation with unannotated image data for a plurality of objects.
15. A non-transitory machine-readable medium having stored thereon instructions for performing a method comprising machine executable code which when executed by at least one machine, causes the machine to:
receive the image data including an object in one or more images corresponding to the image data;
generate, via a neural network based object localized framework, a first set of annotation data including a first bounding box for the object using a first bounding box operation;
receiving feedback on the first set of annotation data; and
determine whether to generate, via the neural network based object localized framework, a second set of annotation data including a second bounding box for the object using a second bounding box operation based on the feedback on the first set of annotation data.
16. The non-transitory machine-readable medium of claim 15 , wherein the first bounding box operation comprises utilizing the neural network to generate convolution layer heatmaps of the object in the one or more images.
17. The non-transitory machine-readable medium of claim 15 , wherein the in response to the feedback being the first bounding box not representing the object in the one or more images:
generating the second bounding box for the object using the second bounding box operation, wherein the second bounding box operation comprises using basic contours in the one or more images from Haar cascades;
receiving, via the input interface, second feedback on the second set of annotation data; and
determining whether to generate a third set of annotation data including a third bounding box for the object using a third bounding box operation based the second feedback on the second set of annotation data.
18. The non-transitory machine-readable medium of claim 17 , wherein in response to the second feedback being second bounding box not representing the object in the one or more images:
generating the third bounding box for the object using the third bounding box operation, wherein the third bounding box operation comprises converting the one or more images using canny edge detection and applying basic contours in the one or more converted images from Haar cascades;
receiving, via the input interface, third feedback on the third set of annotation data; and determining whether to generate a fourth set of annotation data including a fourth bounding box for the object using a fourth bounding box operation based the third feedback on the third set of annotation data.
19. The non-transitory machine-readable medium of claim 18 , wherein in response to the third feedback being the third bounding box not representing the object in the one or more images:
generating the fourth bounding box for the object using the fourth bounding box operation, wherein the fourth bounding box operation comprises applying image erosion to the one or more images and highlighting margins among the at least one object in the one or more images after the image erosion;
receiving, via the input interface, fourth feedback on the fourth set of annotation data; and determining whether to output the fourth set of annotation data including the fourth bounding box based on the fourth feedback.
20. The non-transitory machine-readable medium of claim 15 , wherein in response to the feedback being the first bounding box representing the object in the one or more images, outputting the first set of annotation data.
21. The non-transitory machine-readable medium of claim 15 , wherein the neural network based object localization framework is trained using an unsupervised learning operation with unannotated image data for a plurality of objects.