JPWO2020194111A5 - - Google Patents

Claims (15)

A method for analyzing an image by computer information processing, the method comprising:
obtaining an input image;
obtaining an annotated image specifying regions of interest in the input image;
inputting the input image into a detection model that generates from the input image an output image displaying a target region;
calculating the error between the output image and the annotated image using a loss function that weights errors within the region of interest more heavily than errors outside the region of interest;
A method comprising updating the detection model to reduce the error. 2. The method of claim 1, wherein weighted cross-entropy is used as the loss function. 3. The method of claim 2, wherein the weighted cross-entropy weights within the target region are set to have larger values than the weights outside the target region. A method according to any one of claims 2 to 3, wherein the weighted cross-entropy weights within the region of interest are set to have a greater value than the weights outside the region of interest. The weighted cross-entropy is
Formula 1:

where X is the set of all pixels i in the input image, C is the set of all classes c, Wc is the weight of the class c, and pic is the A method according to any one of claims 2 to 4, wherein q is the value of said class c at pixel i and qic is said value of said class c at said pixel i in said output image. the detection model comprises one or more convolutional layers, one or more pooling layers, one or more deconvolution layers, and one or more batch normalization layers between input and output; The method according to any one of claims 1-5 . wherein the plurality of batch normalization layers comprises a first pass including a portion of the plurality of convolution layers and the plurality of pooling layers and a second pass including the other plurality of convolution layers and the plurality of deconvolution layers; 7. The method of claim 6, wherein each is arranged for a predetermined number of layers contained in at least one of the paths. obtaining at least one coordinate within the input image;
A method according to any one of claims 1 to 7, further comprising specifying said region of interest according to said at least one coordinate. 9. The method of claim 8, wherein said specifying said region of interest according to said at least one coordinate comprises specifying a predefined extent as said region of interest using said at least one coordinate as a reference. 10. Any one of claims 8 to 9, wherein said specifying said target area according to said at least one coordinate comprises specifying, as said target area, a range having a texture similar to the texture at said at least one coordinate. The method according to item 1 . 11. The method according to any one of claims 8 to 10, wherein said specifying said target area according to said at least one coordinate comprises specifying a range enclosed by a set of coordinates as said target area. How . displaying the specified region of interest;
12. The method of any one of claims 8 to 11, further comprising receiving user input to modify the target area while the target area is displayed. Apparatus for implementing the method according to any one of claims 1 to 12 as means by means of computer hardware . A computer program that causes a computer to perform the method according to any one of claims 1 to 12 . 15. A storage medium having recorded thereon the computer program according to claim 14 .