Training method and system of objects detection model based on adaptive annotation design

A training method of object detection model, comprising:

• providing an input image to the object detection model, such that the object detection model is enabled to generate an output image according to the input image;
• calculating a total classification loss value by a loss calculation module according to the output image and a solution image;
• calculating a loss value by the loss calculation module according to the total classification loss value;
• and
• transmitting the loss value to the object detection model by the loss calculation module,
• wherein the total classification loss value is calculated according to a plurality of classification losses corresponding to a plurality of object types;
• each of the classification losses corresponding to each object type is calculated according to a first parameter, a second parameter and a third parameter;
• the first parameter corresponds to a penalty mark determined according to whether one or more annotation boxes belonging to the object type in the output image box-select one or more objects which should be determined as the object type;
• the second parameter corresponds to a threshold value for determining whether the one or more annotation boxes belonging to the object type in the output image box-select one or more objects belonging to the object type;
• and the third parameter corresponds to a mark weighted value determined according to whether the one or more annotation boxes belonging to an object type in the output image box-select one or more objects not belonging to the object type.

A training system of object detection model, comprising:

• an object detection model, comprising a first sub-model and a second sub-model, wherein the first sub-model is configured to generate a first output image according to an input image;
• and the second sub-model is configured to generate a second output image according to the first output image;
• a first loss calculation module coupled to the first sub-model and configured to calculate a first total classification loss value according to the first output image and a solution image, calculate a first loss value according to the first total classification loss value, and transmit the first loss value to the first sub-model;
• and
• a second loss calculation module coupled to the second sub-model and configured to calculate a second total classification loss value according to the second output image and the solution image, calculate a second loss value according to the second total classification loss value, and transmit the second loss value to the second sub-model,
• wherein the first total classification loss value is calculated according to a plurality of first classification losses corresponding to a plurality of object types;
• each of the first classification losses corresponding to each object type is calculated according to a first parameter and a second parameter;
• the second total classification loss value is calculated according to a plurality of second classification losses corresponding to the object types;
• each of the second classification losses corresponding to each object type is calculated according to a third parameter;
• the first parameter corresponds to a penalty mark determined according to whether one or more annotation boxes belonging to an object type in the first output image box-select one or more objects which should be determined as the object type;
• the second parameter corresponds to a threshold value for determining whether the one or more annotation boxes belonging to the object type in the first output image box-select one or more objects belonging to the object type;
• and the third parameter corresponds to a mark weighted value determined according to whether the one or more annotation boxes belonging to the object type in the second output image box-select one or more objects not belonging to the object type.

A training method of object detection model, comprising:

• providing a first sub-model an input image to the object detection model, such that the first sub-model is enabled to generate an output image according to the input image;
• calculating a first total classification loss value by a first loss calculation module according to a first output image and a solution image;
• calculating a first loss value by the first loss calculation module according to the first total classification loss value;
• transmitting the first loss value to the first sub-model by the first loss calculation module;
• generating a second output image by a second sub-model of the object detection model according to the first output image;
• calculating a second total classification loss value by a second loss calculation module according to the second output image and the solution image;
• calculating a second loss value by the second loss calculation module according to the second total classification loss value;
• and
• transmitting the second loss value to the second sub-model by the second loss calculation module,
• wherein the first total classification loss value is calculated according to a plurality of first classification losses corresponding to a plurality of object types;
• each of the first classification losses corresponding to each of the object types is calculated according to a first parameter and a second parameter;
• the second total classification loss value is calculated according to a plurality of second classification losses corresponding to the object types;
• each of the second classification losses corresponding to each object type is calculated according to a third parameter;
• the first parameter corresponds to a penalty mark determined according to whether one or more annotation boxes belonging to an object type in the first output image box-select one or more objects which should be determined as the object type;
• the second parameter corresponds to a threshold value for determining whether the one or more annotation boxes belonging to the object type in the first output image box-select one or more objects belonging to the object type;
• and the third parameter corresponds to a markweighted value determined according to whether the one or more annotation boxes belonging to the object type in the second output image box-select one or more objects not belonging to the object type.