Methods and apparatuses for auto segmentation using bounding box
Provided are a method and an apparatus for auto segmentation using a bounding box. A method for auto segmentation using a bounding box according to one embodiment of the present disclosure comprises receiving a first object image including an object labeled with a bounding box, which is a pre-learning target, learning a segmentation model by classifying an object and a background from the bounding box of the received first object image, and segmenting an object from a second object image, which is an identification target, using the learned segmentation model.
1 . A method for auto segmentation executed by an apparatus for auto segmentation, the method comprising:
receiving a first object image including an object labeled with a bounding box, which is a pre-learning target;
learning a segmentation model by classifying an object and a background from the bounding box of the received first object image; and
segmenting an object from a second object image, which is an identification target, using the learned segmentation model,
wherein the learning a segmentation model calculates a mask loss by summing a first loss calculated by using a mask and a bounding box predicted in the first object image and a second loss calculated by using a mask predicted in the first object image and a color similarity map between individual pixels and their neighboring pixels within the bounding box and learns the segmentation model using the calculated mask loss.
2 . The method of claim 1 , wherein the learning a segmentation model classifies an object and a background using a color similarity map in a bounding box of the received first object image and learns the segmentation model through the classification of the object and the background.
3 . The method of claim 1 , wherein the learning a segmentation model learns the segmentation model by determining whether a pixel located in the bounding box of the received first object image belongs to one of objects to be trained or the background.
4 . The method of claim 1 , wherein the learning a segmentation model calculates a first loss so that the prediction mask is restricted to stay within the bounding box.
5 . The method of claim 1 , wherein the learning a segmentation model calculates a second loss so that an area occupied by the prediction mask contains the minimum of a background area and the maximum of an object area.
6 . The method of claim 1 , further including:
performing auto-labeling in a manner of re-training through user inspection for a bounding box exceeding a preset prediction error value.
7 . The method of claim 1 , wherein the identifying an object from a second object image identifies an object from the second object image, which is an identification target, using the learned segmentation model and a pre-learned multimodal model.
8 . An apparatus for auto segmentation using a bounding box comprising:
a memory storing one or more programs; and
a processor executing the stored one or more programs, wherein the processor is configured to:
receive a first object image including an object labeled with a bounding box, which is a pre-learning target,
learn a segmentation model by classifying an object and a background from the bounding box of the received first object image, and
segment an object from a second object image, which is an identification target, using the learned segmentation model,
wherein the processor calculates a mask loss by summing a first loss calculated by using a mask and a bounding box predicted in the first object image and a second loss calculated by using a mask predicted in the first object image and a color similarity map between individual pixels and their neighboring pixels within the bounding box and learn the segmentation model using the calculated mask loss.
9 . The apparatus of claim 8 , wherein the processor classifies an object and a background using a color similarity map in a bounding box of the received first object image and learns the segmentation model through the classification of the object and the background.
10 . The apparatus of claim 8 , wherein the processor learns the segmentation model by determining whether a pixel located in the bounding box of the received first object image belongs to one of objects to be trained or the background.
11 . The apparatus of claim 8 , wherein the processor calculates a first loss so that the prediction mask is restricted to stay within the bounding box.
12 . The apparatus of claim 8 , wherein the processor calculates a second loss so that an area occupied by the prediction mask contains the minimum of a background area and the maximum of an object area.
13 . The apparatus of claim 8 , wherein the processor performs auto-labeling in a manner of re-training through user inspection for a bounding box exceeding a preset prediction error value.
14 . The apparatus of claim 8 , wherein the processor identifies an object from the second object image, which is an identification target, using the learned segmentation model and a pre-learned multimodal model.
15 . The apparatus of claim 8 comprising a database storing the first object image including the object labeled with the bounding box, which is the pre-learning target, wherein the processor is configured to receive the first object image from the database.