Patent
Atlas literature
Patent
US 10,373,026Patent drawings and their descriptions. Click a drawing to enlarge it.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A method of learning for deriving one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps derived from one or more real images, by using GAN(Generative Adversarial Network) including a generating network and a discriminating networ k, comprising steps of (a) a lea rn ing device, if at least one input image which is one of the virtual images is acquired, instructing the generating network to apply one or more convolutional operations to the input image, to thereby generate at least one output feature map, whose characteristics are same as or similar to those of the real feature maps; and (b) the learning device, if at least one evaluation score, corresponding to the output feature map, generated by the discriminating network is acquired, instructing a first loss unit to generate at least one first loss by referring to the evaluation score, and learning parameters of the generating network by backpropagating the first loss.
The method of Claim 1, wherein the lea rn ing device instructs the discriminating network, capable of determining whether its own inputted feature map is one of the real feature maps or one of the virtual feature maps, to generate at least one probability of the output feature map being real or fake, to thereby generate the evaluation score.
The method of Claim 1, wherein the lea rn ing device instructs an object detection network to generate each of one or more class scores corresponding to each of one or more objects included in the input image by referring to the output feature map.
Amethod of testing for deriving one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps for testing derived from one or more real images, by using GAN(GenerativeAdversarial Network) including a generating network and a discriminating network, comprising a step of on condition that (1) a lea rn ing device has instructed the generating network to apply one or more convolutional operations to at least one training image which is one of the virtual images, to thereby generate at least one output feature map for training whose characteristics are same as or similar to those of one or more real feature maps for training and (2) the lea rn ing device has instructed a first loss unit to generate at least one first loss by referring to at least one evaluation score, corresponding to the output feature map for training generated by the discriminating network, and learning parameters of the generating network by backpropagating the first loss; a testing device, if at least one test image which is one of the virtual images is acquired, instructing the generating network to apply said one or more convolutional operations to the test image, to thereby generate at least one output feature map for testing whose characteristics are same as or similar to those of the real feature maps for testing.
The method of Claim 8, wherein an object detection network detects one or more objects included in the test image by referring to the output feature map for testing.
A learning device for deriving one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps derived from one or more real images, by using GAN(Generative Adversarial Network) including a generating network and a discriminating network, comprising: at least one memo r y that stores instructions; and at least one processor configured to execute the instructions to perform processes of (1) if at least one input image which is one of the virtual images is acquired, instructing the generating network to apply one or more convolutional operations to the input image, to thereby generate at least one output feature map, whose characteristics are same as or similar to those of the real feature maps, and (I), if at least one evaluation score, corresponding to the output feature map, generated by the discriminating network is acquired, instructing a first loss unit to generate at least one first loss by referring to the evaluation score, and lea rn ing parameters of the generating network by backpropagating the first loss.
The learning device of Claim 11, wherein the processor instructs the discriminating network, capable of determining whether its own inputted feature map is one of the real feature maps or one of the virtual feature maps, to generate at least one probability of the output feature map being real or fake, to thereby generate the evaluation score.
The learning device of Claim 11, wherein the processor instructs an object detection network to generate each of one or more class scores corresponding to each of one or more objects included in the input image by referring to the output feature map.
Atesting device for derivin g one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps for testing derived from one or more real images, by using GAN(Generative Adversarial Network) including a generating network and a discriminating network, comprisi ng: at least one memory that stores instructions; and at least one processor, on condition that (1) a lea rn ing device has inst ruc ted the generating network to apply one or more convolutional operations to at least one training image which is one of the virtual images, to thereby generate at least one output feature map for training whose characteristics are same as or similar to those of one or more real feature maps for training and (2) the lea rn ing device has instructed a first loss unit to generate at least one first loss by referring to at least one evaluation score, corresponding to the output feature map for training generated by the discriminating network, and lea rn ing parameters of the generating network by backpropagating the first loss; configured to execute the instructions to: perform a process of, if at least one test image which is one of the virtual images is acquired, instructing the generating network to apply said one or more convolutional operations to the test image, to thereby generate at least one output feature map for testing whose characteristics are same as or similar to those of the real feature maps for testing.
The testing device of Claim 18, wherein an object detection network detects one or more objects included in the test image by referring to the output feature map for testing.
Patent
Atlas literature
Patent
US 10,373,026Patent drawings and their descriptions. Click a drawing to enlarge it.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A method of learning for deriving one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps derived from one or more real images, by using GAN(Generative Adversarial Network) including a generating network and a discriminating networ k, comprising steps of (a) a lea rn ing device, if at least one input image which is one of the virtual images is acquired, instructing the generating network to apply one or more convolutional operations to the input image, to thereby generate at least one output feature map, whose characteristics are same as or similar to those of the real feature maps; and (b) the learning device, if at least one evaluation score, corresponding to the output feature map, generated by the discriminating network is acquired, instructing a first loss unit to generate at least one first loss by referring to the evaluation score, and learning parameters of the generating network by backpropagating the first loss.
The method of Claim 1, wherein the lea rn ing device instructs the discriminating network, capable of determining whether its own inputted feature map is one of the real feature maps or one of the virtual feature maps, to generate at least one probability of the output feature map being real or fake, to thereby generate the evaluation score.
The method of Claim 1, wherein the lea rn ing device instructs an object detection network to generate each of one or more class scores corresponding to each of one or more objects included in the input image by referring to the output feature map.
Amethod of testing for deriving one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps for testing derived from one or more real images, by using GAN(GenerativeAdversarial Network) including a generating network and a discriminating network, comprising a step of on condition that (1) a lea rn ing device has instructed the generating network to apply one or more convolutional operations to at least one training image which is one of the virtual images, to thereby generate at least one output feature map for training whose characteristics are same as or similar to those of one or more real feature maps for training and (2) the lea rn ing device has instructed a first loss unit to generate at least one first loss by referring to at least one evaluation score, corresponding to the output feature map for training generated by the discriminating network, and learning parameters of the generating network by backpropagating the first loss; a testing device, if at least one test image which is one of the virtual images is acquired, instructing the generating network to apply said one or more convolutional operations to the test image, to thereby generate at least one output feature map for testing whose characteristics are same as or similar to those of the real feature maps for testing.
The method of Claim 8, wherein an object detection network detects one or more objects included in the test image by referring to the output feature map for testing.
A learning device for deriving one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps derived from one or more real images, by using GAN(Generative Adversarial Network) including a generating network and a discriminating network, comprising: at least one memo r y that stores instructions; and at least one processor configured to execute the instructions to perform processes of (1) if at least one input image which is one of the virtual images is acquired, instructing the generating network to apply one or more convolutional operations to the input image, to thereby generate at least one output feature map, whose characteristics are same as or similar to those of the real feature maps, and (I), if at least one evaluation score, corresponding to the output feature map, generated by the discriminating network is acquired, instructing a first loss unit to generate at least one first loss by referring to the evaluation score, and lea rn ing parameters of the generating network by backpropagating the first loss.
The learning device of Claim 11, wherein the processor instructs the discriminating network, capable of determining whether its own inputted feature map is one of the real feature maps or one of the virtual feature maps, to generate at least one probability of the output feature map being real or fake, to thereby generate the evaluation score.
The learning device of Claim 11, wherein the processor instructs an object detection network to generate each of one or more class scores corresponding to each of one or more objects included in the input image by referring to the output feature map.
Atesting device for derivin g one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps for testing derived from one or more real images, by using GAN(Generative Adversarial Network) including a generating network and a discriminating network, comprisi ng: at least one memory that stores instructions; and at least one processor, on condition that (1) a lea rn ing device has inst ruc ted the generating network to apply one or more convolutional operations to at least one training image which is one of the virtual images, to thereby generate at least one output feature map for training whose characteristics are same as or similar to those of one or more real feature maps for training and (2) the lea rn ing device has instructed a first loss unit to generate at least one first loss by referring to at least one evaluation score, corresponding to the output feature map for training generated by the discriminating network, and lea rn ing parameters of the generating network by backpropagating the first loss; configured to execute the instructions to: perform a process of, if at least one test image which is one of the virtual images is acquired, instructing the generating network to apply said one or more convolutional operations to the test image, to thereby generate at least one output feature map for testing whose characteristics are same as or similar to those of the real feature maps for testing.
The testing device of Claim 18, wherein an object detection network detects one or more objects included in the test image by referring to the output feature map for testing.
Patent
Atlas literature
Patent
US 10,373,026Patent drawings and their descriptions. Click a drawing to enlarge it.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A method of learning for deriving one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps derived from one or more real images, by using GAN(Generative Adversarial Network) including a generating network and a discriminating networ k, comprising steps of (a) a lea rn ing device, if at least one input image which is one of the virtual images is acquired, instructing the generating network to apply one or more convolutional operations to the input image, to thereby generate at least one output feature map, whose characteristics are same as or similar to those of the real feature maps; and (b) the learning device, if at least one evaluation score, corresponding to the output feature map, generated by the discriminating network is acquired, instructing a first loss unit to generate at least one first loss by referring to the evaluation score, and learning parameters of the generating network by backpropagating the first loss.
The method of Claim 1, wherein the lea rn ing device instructs the discriminating network, capable of determining whether its own inputted feature map is one of the real feature maps or one of the virtual feature maps, to generate at least one probability of the output feature map being real or fake, to thereby generate the evaluation score.
The method of Claim 1, wherein the lea rn ing device instructs an object detection network to generate each of one or more class scores corresponding to each of one or more objects included in the input image by referring to the output feature map.
Amethod of testing for deriving one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps for testing derived from one or more real images, by using GAN(GenerativeAdversarial Network) including a generating network and a discriminating network, comprising a step of on condition that (1) a lea rn ing device has instructed the generating network to apply one or more convolutional operations to at least one training image which is one of the virtual images, to thereby generate at least one output feature map for training whose characteristics are same as or similar to those of one or more real feature maps for training and (2) the lea rn ing device has instructed a first loss unit to generate at least one first loss by referring to at least one evaluation score, corresponding to the output feature map for training generated by the discriminating network, and learning parameters of the generating network by backpropagating the first loss; a testing device, if at least one test image which is one of the virtual images is acquired, instructing the generating network to apply said one or more convolutional operations to the test image, to thereby generate at least one output feature map for testing whose characteristics are same as or similar to those of the real feature maps for testing.
The method of Claim 8, wherein an object detection network detects one or more objects included in the test image by referring to the output feature map for testing.
A learning device for deriving one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps derived from one or more real images, by using GAN(Generative Adversarial Network) including a generating network and a discriminating network, comprising: at least one memo r y that stores instructions; and at least one processor configured to execute the instructions to perform processes of (1) if at least one input image which is one of the virtual images is acquired, instructing the generating network to apply one or more convolutional operations to the input image, to thereby generate at least one output feature map, whose characteristics are same as or similar to those of the real feature maps, and (I), if at least one evaluation score, corresponding to the output feature map, generated by the discriminating network is acquired, instructing a first loss unit to generate at least one first loss by referring to the evaluation score, and lea rn ing parameters of the generating network by backpropagating the first loss.
The learning device of Claim 11, wherein the processor instructs the discriminating network, capable of determining whether its own inputted feature map is one of the real feature maps or one of the virtual feature maps, to generate at least one probability of the output feature map being real or fake, to thereby generate the evaluation score.
The learning device of Claim 11, wherein the processor instructs an object detection network to generate each of one or more class scores corresponding to each of one or more objects included in the input image by referring to the output feature map.
Atesting device for derivin g one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps for testing derived from one or more real images, by using GAN(Generative Adversarial Network) including a generating network and a discriminating network, comprisi ng: at least one memory that stores instructions; and at least one processor, on condition that (1) a lea rn ing device has inst ruc ted the generating network to apply one or more convolutional operations to at least one training image which is one of the virtual images, to thereby generate at least one output feature map for training whose characteristics are same as or similar to those of one or more real feature maps for training and (2) the lea rn ing device has instructed a first loss unit to generate at least one first loss by referring to at least one evaluation score, corresponding to the output feature map for training generated by the discriminating network, and lea rn ing parameters of the generating network by backpropagating the first loss; configured to execute the instructions to: perform a process of, if at least one test image which is one of the virtual images is acquired, instructing the generating network to apply said one or more convolutional operations to the test image, to thereby generate at least one output feature map for testing whose characteristics are same as or similar to those of the real feature maps for testing.
The testing device of Claim 18, wherein an object detection network detects one or more objects included in the test image by referring to the output feature map for testing.
Patent
Atlas literature
Patent
US 10,373,026Patent drawings and their descriptions. Click a drawing to enlarge it.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A method of learning for deriving one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps derived from one or more real images, by using GAN(Generative Adversarial Network) including a generating network and a discriminating networ k, comprising steps of (a) a lea rn ing device, if at least one input image which is one of the virtual images is acquired, instructing the generating network to apply one or more convolutional operations to the input image, to thereby generate at least one output feature map, whose characteristics are same as or similar to those of the real feature maps; and (b) the learning device, if at least one evaluation score, corresponding to the output feature map, generated by the discriminating network is acquired, instructing a first loss unit to generate at least one first loss by referring to the evaluation score, and learning parameters of the generating network by backpropagating the first loss.
The method of Claim 1, wherein the lea rn ing device instructs the discriminating network, capable of determining whether its own inputted feature map is one of the real feature maps or one of the virtual feature maps, to generate at least one probability of the output feature map being real or fake, to thereby generate the evaluation score.
The method of Claim 1, wherein the lea rn ing device instructs an object detection network to generate each of one or more class scores corresponding to each of one or more objects included in the input image by referring to the output feature map.
Amethod of testing for deriving one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps for testing derived from one or more real images, by using GAN(GenerativeAdversarial Network) including a generating network and a discriminating network, comprising a step of on condition that (1) a lea rn ing device has instructed the generating network to apply one or more convolutional operations to at least one training image which is one of the virtual images, to thereby generate at least one output feature map for training whose characteristics are same as or similar to those of one or more real feature maps for training and (2) the lea rn ing device has instructed a first loss unit to generate at least one first loss by referring to at least one evaluation score, corresponding to the output feature map for training generated by the discriminating network, and learning parameters of the generating network by backpropagating the first loss; a testing device, if at least one test image which is one of the virtual images is acquired, instructing the generating network to apply said one or more convolutional operations to the test image, to thereby generate at least one output feature map for testing whose characteristics are same as or similar to those of the real feature maps for testing.
The method of Claim 8, wherein an object detection network detects one or more objects included in the test image by referring to the output feature map for testing.
A learning device for deriving one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps derived from one or more real images, by using GAN(Generative Adversarial Network) including a generating network and a discriminating network, comprising: at least one memo r y that stores instructions; and at least one processor configured to execute the instructions to perform processes of (1) if at least one input image which is one of the virtual images is acquired, instructing the generating network to apply one or more convolutional operations to the input image, to thereby generate at least one output feature map, whose characteristics are same as or similar to those of the real feature maps, and (I), if at least one evaluation score, corresponding to the output feature map, generated by the discriminating network is acquired, instructing a first loss unit to generate at least one first loss by referring to the evaluation score, and lea rn ing parameters of the generating network by backpropagating the first loss.
The learning device of Claim 11, wherein the processor instructs the discriminating network, capable of determining whether its own inputted feature map is one of the real feature maps or one of the virtual feature maps, to generate at least one probability of the output feature map being real or fake, to thereby generate the evaluation score.
The learning device of Claim 11, wherein the processor instructs an object detection network to generate each of one or more class scores corresponding to each of one or more objects included in the input image by referring to the output feature map.
Atesting device for derivin g one or more virtual feature maps from one or more virtual images, whose one or more characteristics are same as or similar to those of one or more real feature maps for testing derived from one or more real images, by using GAN(Generative Adversarial Network) including a generating network and a discriminating network, comprisi ng: at least one memory that stores instructions; and at least one processor, on condition that (1) a lea rn ing device has inst ruc ted the generating network to apply one or more convolutional operations to at least one training image which is one of the virtual images, to thereby generate at least one output feature map for training whose characteristics are same as or similar to those of one or more real feature maps for training and (2) the lea rn ing device has instructed a first loss unit to generate at least one first loss by referring to at least one evaluation score, corresponding to the output feature map for training generated by the discriminating network, and lea rn ing parameters of the generating network by backpropagating the first loss; configured to execute the instructions to: perform a process of, if at least one test image which is one of the virtual images is acquired, instructing the generating network to apply said one or more convolutional operations to the test image, to thereby generate at least one output feature map for testing whose characteristics are same as or similar to those of the real feature maps for testing.
The testing device of Claim 18, wherein an object detection network detects one or more objects included in the test image by referring to the output feature map for testing.
