Synthesizing training samples for object recognition
An enhanced training sample set containing new synthesized training images that are artificially generated from an original training sample set is provided to satisfactorily increase the accuracy of an object recognition system. The original sample set is artificially augmented by introducing one or more variations to the original images with little to no human input. There are a large number of possible variations that can be introduced to the original images, such as varying the image's position, orientation, and/or appearance and varying an object's context, scale, and/or rotation. Because there are computational constraints on the amount of training samples that can be processed by object recognition systems, one or more variations that will lead to a satisfactory increase in the accuracy of the object recognition performance are identified and introduced to the original images.
1. A method comprising:
generating synthesized training samples by applying one or more object-based transform models to original training samples, the transform models including an object-based transform model that relocates an object from a source image to a target image selected based on presence of one or more co-occurring objects between the source image and the target image;
validating the one or more object-based transform models based on image processing accuracy for an object class, based on the synthesized training samples;
selecting one or more object-based transform models, the selected object-based transform models having a satisfactorily increased image processing accuracy for the object class wherein the one or more object-based transform models includes an object-based transform model that relocates an object from a source image to a different image region within a background of a target image; and
determining co-occurrence statistics of a plurality of co-occurring objects in the source image and selecting the target image from a plurality of images based on co-occurrence statistics of co-occurring objects present in the target image.
2. The method of claim 1 wherein the target image is selected based on a similarity of background value between the source image and the target image.
3. The method of claim 1 wherein the one or more object-based transform models includes an object-based transform model that scales an object.
4. The method of claim 1 wherein the one or more object-based transform models includes an object-based transform model that rotates an object.
5. One or more computer-readable storage media, wherein the one or more computer readable storage media is not a carrier wave, encoding computer-executable instructions for executing on a computer system a computer process, the computer process comprising:
generating synthesized training samples by applying one or more object-based transform models to original training samples, the transform models including an object-based transform model that relocates an object from a source image to a target image selected based on presence of one or more co-occurring objects between the source image and the target image, wherein the one or more object-based transform models include an object-based transform model that relocates an object from a source image to a different image region within a background of the target image and wherein the target image is selected based on co-occurrence statistics of one or more co-occurring objects in the target image.
6. The one or more computer-readable storage media of claim 5 wherein the target image is selected based on a similarity of background value between the source image and the target image.
7. A system comprising:
a synthesizer module implemented by a processor to generate synthesized training samples by applying one or more object-based transform models to original training samples, the transform models including an object-based transform model that relocates an object from a source image to a target image selected based on presence of one or more co-occurring objects between the source image and the target image, wherein the one or more object-based transforms models include an object-based transform model that relocates an object from a source image to a different image region within a background of the target image and wherein the synthesizer module is further configured to determine co-occurrence statistics for a plurality of co-occurring objects in the source image and to select the target image from a plurality of images based on co-occurrence statistics of co-occurring objects present in the target image.
8. The system of claim 7 wherein the target image is selected based on a similarity of background value between the source image and the target image.