Object verification/recognition with limited input
Systems and methods for object recognition with limited input are disclosed herein. An example method includes updating a neural network trained to perform object recognition on a first rendition of an object, so that the neural network performs object recognition on a second rendition of the object, using a limited set of input images. The method includes receiving a limited set of model images of the second rendition of the object, accessing a corresponding image mapping, and generating a large number of training images from the limited set, where image mappings include geometric, illumination, and/or obscuration transformations. The neural network is then trained, from this initial small set, to classify the second rendition of the object.
1. A computer-implemented method for updating a neural network trained to recognize a first rendition of an object, the method comprising:
receiving, at one or more processors, a plurality of model images of a second rendition of the object, the model images of second rendition of the object satisfy model image conditions;
accessing, at the one or more processors, predetermined image mapping between training images captured for the first rendition of the object and model images of the first rendition of the object, wherein the model images of the first rendition of the object satisfy the model image conditions;
applying, at the one or more processors, the predetermined image mapping to the plurality of model images of the second rendition of the object to generate training images of the second rendition of the object: and
performing, at the one or more processors, updated training on the neural network using the training images of the second rendition of the object.
2. The computer-implemented method of claim 1 , wherein performing updated training on the neural network using the plurality of training images of the second rendition of the object comprises:
adding the training images of the second rendition of the object to the training images captured for the first rendition of the object; and
training the neural network to classify subsequent images as corresponding to either one of the second rendition of the object or the first rendition of the object.
3. The computer-implemented method of claim 1 , wherein performing updated training on the neural network using the plurality of training images of the second rendition of the object comprises:
replacing the training images captured for the first rendition of the object with the training images of the second rendition of the object; and
training the neural network to classify subsequent images as corresponding to the second rendition of the object.
4. The computer-implemented method of claim 1 , wherein predetermined image mapping between the training images captured for the first rendition of the object and the model images of the first rendition of the object comprises geometric transformations.
5. The computer-implemented method of claim 1 , wherein predetermined image mapping between the training images captured for the first rendition of the object and the model images of the first rendition of the object comprises illumination transformations.
6. The computer-implemented method of claim 1 , wherein predetermined image mapping between the training images captured for the first rendition of the object and the model images of the first rendition of the object comprises obscuration transformations.
7. The computer-implemented method of claim 1 , further comprising, in addition to applying the predetermined image mapping to the plurality of model images of the second rendition of the object applying at least one of a geometric transformation, an illumination transformation, and an obscuration transformation to the plurality of model images of the second rendition of the object to generate the training images of the second rendition of the object.
8. The computer-implemented method of claim 1 , further comprising:
receiving, at the one or more processors, subsequent image data from an imager assembly;
analyzing, in the neural network, the subsequent image data and classifying the subsequent image data as corresponding to the second rendition of the object.
9. A system comprising:
an imager assembly configured to capture a plurality of model images of an object; and
a processor and memory storing instructions that, when executed, cause the processor to:
access a predetermined image mapping between training images and model images of a previously-trained object, wherein the object is a rendition of the previously trained object;
apply the predetermined image mapping to the plurality of model images of the object to generate training images of the object: and
perform updated training on a neural network using generated training images of the object.
10. The system of claim 8 , wherein the memory storing further instructions that, when executed, cause the processor to:
add the training images of the object to the training images of the previously-trained object; and
train the neural network to classify subsequent images as corresponding to either one of the object or the previously-trained object.
11. The system of claim 8 , wherein the memory storing further instructions that, when executed, cause the processor to:
replace the training images of the previously-trained object with the training images of the object; and
train the neural network to classify subsequent images as corresponding to the object, and not to the previously-trained object.
12. The system of claim 8 , wherein the predetermined image mapping comprises geometric transformations.
13. The system of claim 8 , wherein the predetermined image mapping comprises illumination transformations.
14. The system of claim 8 , wherein the predetermined image mapping comprises obscuration transformations.
15. The system of claim 8 , wherein predetermined image mapping comprises a geometric transformation, an illumination transformation, and an obscuration transformation.
16. The system of claim 8 , wherein the memory storing further instructions that, when executed, cause the processor to:
receive subsequent image data from an imager assembly; and
analyze, in the neural network, the subsequent image data and classify the subsequent image data as corresponding to the second rendition of the object.