An image processing apparatus includes an inference part which is capable of inference by using a first inference model for finding of a specific target object and by using a second inference model for discrimination about the specific target object and a control unit to which a first picked-up image obtained under a first image pickup condition and a second picked-up image obtained under a second image pickup condition different from the first image pickup condition can be inputted and which performs control such that in a case where the first picked-up image is inputted, the inference part is caused to execute inference by using the first inference model, and in a case where the second picked-up image is inputted, the inference part is caused to execute inference by using the second inference model.