Abstract: The present invention relates to a neural network for object segmentation. A method for object segmentation using a trained neural network according to an embodiment of the present invention includes receiving a segmentation target image, splitting the target image into unit images having a predetermined size, outputting a unit activation map for the split unit image as a first segmentation result by using the neural network, merging the unit activation map, and outputting a second segmentation result according to the merged unit activation map, in which the neural network is trained by mutually using the entire activation map for the target image and the merged activation map. According to the present invention, it is possible to generate a more accurate object segmentation result, and generate a label using the segmentation result and use the generated label for training of a neural network.

Abstract: A composition containing a Dendropanax morbifera LEV. extract, of the present invention, has excellent activities of inhibiting the oxidation of A2E, inhibiting oxidized A2E-induced death of retinal pigmented epithelial cells, and inhibiting the accumulation of drusen, and thus can be used for preventing eye damage. In addition, a composition containing a Dendropanax morbifera LEV. extract, of the present invention, can be developed even into a therapeutic agent for various eye diseases and a health functional food.

Abstract: An anode for a lithium secondary battery includes an anode current collector, and an anode active material layer formed on at least one surface of the anode current collector. The anode active material layer includes a carbon-based active material, a first silicon-based active material including a carbon-silicon composite active material, and a second silicon-based active material including a silicon oxide (SiOx, 0<x<2). A content of the first silicon-based active material is in a range from 2 wt % to 40 wt % based on a total weight of the anode active material layer.

Abstract: The present application discloses a method for treating a protein deficiency in the central nervous system of a subject in need thereof, comprising systemically administering to the subject a therapeutically effective dose of a fusion polypeptide comprising the first protein, wherein the fusion polypeptide comprises: (a) the first protein; (b) a second protein that provides extended circulation-lifetime in vivo and (c) blood brain barrier crossing facilitating peptide; wherein the fusion polypeptide crosses the blood brain barrier (BBB).

Abstract: A display device includes a data line, a vertical transmission line and a horizontal transmission line. The data line extends in a first direction and is arranged in a second direction on a display area divided into upper and lower display areas, and is connected to pixels. A vertical transmission line extends in the first direction on the display area. The vertical transmission line receives a ground power supply voltage from an upper side of the display area to transfer the ground power supply voltage to the pixels, and receives a data voltage from a lower side of the display area. A horizontal transmission line extends in the second direction on the display area and is arranged in the first direction. The horizontal transmission line is connected to the vertical transmission line and the data line on the lower display area to transfer the data voltage to the data line.

Abstract: An artificial turf structure includes a protective layer, a buffer layer disposed on a lower surface of the protective layer and having a three-dimensional structure comprising a front surface layer, an intermediate layer and a rear surface layer, a bubble layer disposed on a lower surface of the buffer layer, a pile unit tufted to the protective layer, the buffer layer and the bubble layer, and a backing layer disposed on a lower surface of the bubble layer and preventing the pile unit from leaving.

Abstract: A lithium secondary battery is disclosed. In some implementations, the lithium secondary battery includes a positive electrode including a positive electrode active material including lithium metal oxide including nickel (Ni) and a negative electrode including a negative electrode active material including a silicon-based active material, wherein the positive electrode active material includes 10 wt % or more of lithium metal oxide in the form of single particles based on a total weight of the positive electrode active material, and the negative electrode active material includes 1 to 15 wt % of the silicon-based active material based on a total weight of the negative electrode active material.

Abstract: An anode for a secondary battery includes an anode current collector, a first anode mixture layer on at least one surface of the anode current collector, and a second anode mixture layer on the first anode mixture layer. The first anode mixture layer includes a first silicon-based active material having a carbon coating layer formed on a surface, and a first conductive material, the second anode mixture layer includes a second silicon-based active material doped with a metal, and a second conductive material, and the anode for a secondary battery exhibits a Radial Breathing Mode (RBM) peak in a Raman spectrum obtained from a surface of the second anode mixture layer. An influence of volume expansion/contraction of a silicon-based active material during battery charging/discharging may be alleviated.

Abstract: Disclosed are a frame transmission method using a selective beamforming and a communication apparatus to perform the frame transmission method. The communication apparatus may determine a beamforming matrix based on classification information in which a plurality of subcarriers used for communication is classified into a plurality of frequency units, may map a long training field (LTF) sequence to the beamforming matrix, and transmit a beamforming training (BF-T) frame including the mapped LTF sequence to a plurality of stations, may receive, from the plurality of stations having receiving the BF-T frame, feedback information generated based on a reception strength of the BF-T frame, and may allocate frequency units to data frames to be transmitted to the plurality of stations based on the feedback information, and transmit the data frames using the allocated frequency units. The reception strength of the BF-T frame may be determined at each station for each frequency unit.

Abstract: An anode for a secondary battery is disclosed. In some implementations, the anode includes an anode current collector, a first anode mixture layer on at least one surface of the anode current collector, and a second anode mixture layer on the first anode mixture layer. The first anode mixture layer includes a first silicon-based active material including a carbon coating layer formed on a surface thereof, and a first conductive material. The second anode mixture layer includes a second silicon-based active material doped with a metal, and a second conductive material. The first conductive material has a Raman R value, greater than or equal to a Raman R value of the second conductive material. According to some implementations, volume expansion/contraction of a silicon-based active material may be alleviated during battery charging/discharging.

Abstract: Disclosed herein is a learning method of a neural network model for flame determination. The learning method of a neural network includes generating a learning image including a fake image generated by combining a real fire image and an arbitrary flame image with a background image; inputting the learning image to a first neural network model and outputting a determination result for whether a flame is present; and updating a weight in a layer extracting features of the learning image from the first neural network model using the determination result. According to the present invention, data of various fire situations may be secured, a performance of the neural network model that determines an occurrence of the fire through the secured data may be increased, and a quality of data for learning may be increased to allow the neural network model itself to predict various situations of fires.

Abstract: Disclosed herein is a learning method of a neural network model for flame determination. The learning method of a neural network includes generating a learning image including a fake image generated by combining a real fire image and an arbitrary flame image with a background image; inputting the learning image to a first neural network model and outputting a determination result for whether a flame is present; and updating a weight in a layer extracting features of the learning image from the first neural network model using the determination result. According to the present invention, data of various fire situations may be secured, a performance of the neural network model that determines an occurrence of the fire through the secured data may be increased, and a quality of data for learning may be increased to allow the neural network model itself to predict various situations of fires.

Abstract: Disclosed herein is a learning method of a neural network model for flame determination. The learning method of a neural network includes generating a learning image including a fake image generated by combining a real fire image and an arbitrary flame image with a background image; inputting the learning image to a first neural network model and outputting a determination result for whether a flame is present; and updating a weight in a layer extracting features of the learning image from the first neural network model using the determination result. According to the present invention, data of various fire situations may be secured, a performance of the neural network model that determines an occurrence of the fire through the secured data may be increased, and a quality of data for learning may be increased to allow the neural network model itself to predict various situations of fires.

Abstract: A backlight device includes: a light source to emit coherent light; an optical path difference generator on the light source, the optical path difference generator including an incident surface and a plurality of light emitting surfaces, the light emitting surfaces being parallel to the incident surface and having different separation distances from the incident surface; a light condenser on the optical path difference generator; a diffuser on the light condenser; and a collimator on the diffuser.

Abstract: The present invention relates to a neural network for object segmentation. A method for object segmentation using a trained neural network according to an embodiment of the present invention includes receiving a segmentation target image, splitting the target image into unit images having a predetermined size, outputting a unit activation map for the split unit image as a first segmentation result by using the neural network, merging the unit activation map, and outputting a second segmentation result according to the merged unit activation map, in which the neural network is trained by mutually using the entire activation map for the target image and the merged activation map. According to the present invention, it is possible to generate a more accurate object segmentation result, and generate a label using the segmentation result and use the generated label for training of a neural network.

Abstract: The present invention relates to a method for training a neural network for object detection. The method includes receiving a detection target image, splitting the detection target image into unit images having a predetermined size, defining an output of the neural network for the split unit images as a first label value, generating a first deformed image by deforming the unit image according to a first rule, and training the neural network by using an output of the neural network for the first deformed image and a loss of the first label value. According to the present invention, it is possible to efficiently train a neural network for detecting an object in a large screen.

Abstract: Disclosed herein is a learning method of a neural network model for flame determination. The learning method of a neural network includes generating a learning image including a fake image generated by combining a real fire image and an arbitrary flame image with a background image; inputting the learning image to a first neural network model and outputting a determination result for whether a flame is present; and updating a weight in a layer extracting features of the learning image from the first neural network model using the determination result. According to the present invention, data of various fire situations may be secured, a performance of the neural network model that determines an occurrence of the fire through the secured data may be increased, and a quality of data for learning may be increased to allow the neural network model itself to predict various situations of fires.

Abstract: Disclosed herein is a learning method of a neural network model for flame determination. The learning method of a neural network includes generating a learning image including a fake image generated by combining a real fire image and an arbitrary flame image with a background image; inputting the learning image to a first neural network model and outputting a determination result for whether a flame is present; and updating a weight in a layer extracting features of the learning image from the first neural network model using the determination result. According to the present invention, data of various fire situations may be secured, a performance of the neural network model that determines an occurrence of the fire through the secured data may be increased, and a quality of data for learning may be increased to allow the neural network model itself to predict various situations of fires.

Abstract: A backlight device includes: a light source to emit coherent light; an optical path difference generator on the light source, the optical path difference generator including an incident surface and a plurality of light emitting surfaces, the light emitting surfaces being parallel to the incident surface and having different separation distances from the incident surface; a light condenser on the optical path difference generator; a diffuser on the light condenser; and a collimator on the diffuser.

Abstract: Disclosed herein is a learning method of a neural network model for flame determination. The learning method of a neural network includes generating a learning image including a fake image generated by combining a real fire image and an arbitrary flame image with a background image; inputting the learning image to a first neural network model and outputting a determination result for whether a flame is present; and updating a weight in a layer extracting features of the learning image from the first neural network model using the determination result. According to the present invention, data of various fire situations may be secured, a performance of the neural network model that determines an occurrence of the fire through the secured data may be increased, and a quality of data for learning may be increased to allow the neural network model itself to predict various situations of fires.