Object detection device, learning method, and recording medium
In an object detection device, a plurality of object detection units output a score indicating probability that a predetermined object exists, for each partial region set to image data inputted. The weight computation unit computes weights for merging the scores outputted by the plurality of object detection units, using weight calculation parameters, based on the image data. The merging unit merges the scores outputted by the plurality of object detection units, for each partial region, with the weights computed by the weight computation unit. The target model object detection unit configured to output a score indicating probability that the predetermined object exists, for each partial region set to the image data. The first loss computation unit computes a first loss indicating a difference of the score of the target model object detection unit from a ground truth label of the image data and the score merged by the merging unit. The first parameter correction unit corrects parameters of the target model object detection unit to reduce the first loss.
1. An object detection device comprising:
a memory storing instructions; and
one or more processors configured to execute the instructions to:
respectively output, by a plurality of object detection units, scores indicating a probability that a predetermined object exists, for each of a plurality of partial regions set to image data that has been input;
compute weights for merging the scores output by the plurality of object detection units for each partial region, using weight calculation parameters, based on the image data;
merge the scores output by the plurality of object detection units, for each partial region, using the computed weights, to generate a merged score for each partial region-;
output, by a target model object detection unit, a score indicating a probability that the predetermined object exists, for each partial region;
compute a first loss indicating a difference of the score of the target model object detection unit from a ground truth label of the image data and the merged score, for each partial region; and
correct parameters of the target model object detection unit to reduce the first loss.
2. The object detection device according to claim 1 , wherein the processor is further configured to:
compute a second loss indicating a difference between the ground truth label and the merged score, for each partial region; and
correct the weight calculation parameters to reduce the second loss.
3. The object detection device according to claim 2 , wherein the processor is configured to correct the weight computation parameters based on the first loss and the second loss.
4. The object detection device according to claim 2 ,
wherein processor is configured to output, by each of the plurality of object detection units and the target model object detection unit, coordinate information of a rectangular region in which the predetermined object exists for each partial region,
wherein the processor is configured to merge the coordinate information with the computed weights,
wherein the processor is configured to compute a first loss indicating a difference of the output of the target model object detection unit from the ground truth label and the merged score-, for each partial region, and
wherein the processor is configured to compute a loss including a difference between the merged coordinate information merged and the ground truth label.
5. The object detection device according to claim 4 , wherein the processor is configured to multiply the coordinate information output by the plurality of object detection units by the computed weights, respectively, and compute an average value based on the multiplied scores.
6. The object detection device according claim 1 ,
wherein the image data includes unlabeled image data that does not have a corresponding ground truth label, and
wherein, for the unlabeled image data, the processor is configured to compute a difference between the score output by the target model object detection unit and the merged score as the first loss, for each partial region.
7. The object detection device according to claim 1 ,
wherein the processor is configured to respectively compute single weights for the image data for the plurality of object detection units, and
wherein the processor is configured to merge the scores output by the plurality of object detection units, for each partial region, using single weights.
8. The object detection device according to claim 1 ,
wherein the processor is configured to respectively compute weights for each of the partial regions for the plurality of object detection units, and
wherein the processor is configured to merge the scores output by the plurality of object detection units, for each partial region, using the weights computed for each partial region.
9. The object detection device according to claim 1 ,
wherein the processor is configured to compute the weights for each of a plurality of classes indicating the predetermined object, and
wherein the processor is configured to merge the scores output by the plurality of object detection units with the weights computed for each class.
10. The object detection device according to claim 1 , wherein the processor is configured to multiply the scores output by the plurality of object detection units by the computed weights, respectively, and compute an average value based on the multiplied scores.
11. A learning method performed by an object detection device and comprising:
respectively outputting, by a plurality of object detection units, scores indicating a probability that a predetermined object exists, for each of a plurality of partial regions set to image data that has been input;
computing weights for merging the scores output by the plurality of object detection units for each partial region, using weight calculation parameters, based on the image data;
merging the scores output by the plurality of object detection units, for each partial region, using the computed weights, to generate a merged score for each partial region;
outputting, by a target model object detection unit, a score indicating a probability that the predetermined object exists, for each partial region;
computing a first loss indicating a difference of the score of the target model object detection unit from a ground truth label of the image data and the merged score, for each partial region; and
correcting parameters of the target model object detection unit to reduce the first loss.
12. A non-transitory computer-readable recording medium storing a program executable by an object detection device, to perform learning processing comprising:
respectively outputting, by a plurality of object detection units, scores indicating a probability that a predetermined object exists, for each of a plurality of partial regions set to image data that has been input;
computing weights for merging the scores output by the plurality of object detection units for each partial region, using weight calculation parameters, based on the image data;
merging the scores output by the plurality of object detection units, for each partial region, using the computed weights, to generate a merged score for each partial region;
outputting, by a target model object detection unit, a score indicating a probability that the predetermined object exists, for each partial region;
computing a first loss indicating a difference of the score of the target model object detection unit from a ground truth label of the image data and the merged score, for each partial region; and
correcting parameters of the target model object detection unit to reduce the first loss.