Abstract: Embodiments of this disclosure provide an image encoding method and apparatus and image decoding method and apparatus. The image encoding includes performing convolutional neural network (CNN) encoding on image data to generate feature vectors or feature maps; quantizing the feature vectors or feature maps to generate discrete symbols to be encoded; and estimating probabilities of the symbols to be encoded by using a multi-scale context model including multiple mask convolution layers of different scales. An entropy encoding of the image data is performed according to the probabilities of the symbols to be encoded.

Abstract: Embodiments of this disclosure provide an image coding apparatus, a probability model generating apparatus and an image decoding apparatus. A processor is to perform feature extraction on an input image to obtain first feature maps of N channels; to perform feature extraction on the input image with a size of the input image being adjusted K times, to respectively obtain second feature maps of N channels; and to concatenate the first feature maps of the K×N channels with the second feature maps of K×N channels to output a concatenated feature maps of channels. Hence, features of images may be accurately extracted and more competitive latent representations may be obtained.

Abstract: Embodiments of this disclosure provide a video frame interpolation apparatus and method. The method includes calculating a bidirectional optical flow between a first frame and a second frame; an performing kernel and weight estimation according to the first frame and the second frame. An adaptive local convolutional kernel is generated by using a convolutional layer and a weight coefficient is generated by using another convolutional layer. A conversion on the first frame and the second frame is performed by using an adaptive conversion layer according to the bidirectional optical flow, the weight coefficient and the adaptive local convolutional kernel, so as to generate a conversion frame. A frame synthesis on the first frame, the second frame and the conversion frame is performed to generate an interpolation frame between the first frame and the second frame.

Abstract: Embodiments of this disclosure provide a training method, an image encoding method, an image decoding method and apparatuses thereof. The image encoding apparatus includes: an image encoder configured to encode input image data to obtain a latent variable; a quantizer configured to perform quantizing processing on the latent variable according to a quantization step to generate a quantized latent variable; and an entropy encoder configured to perform entropy coding on the quantized latent variable by using an entropy model to form a bit stream.

Abstract: Embodiments of this disclosure provide an image coding method and apparatus and an image compression system. The image coding apparatus includes a memory and a processor. The processor is configured to perform feature extraction on an input image to obtain feature maps of N channels; assign a weight to a feature map of each channel among the N channels; perform down-dimension processing on weighted feature maps processed in association with the N channels, to obtain feature maps of M channels and output the feature maps of M channels, M being smaller than N. Hence, by multiplying different feature maps by a weight to obtain corresponding importance and then performing down-dimension processing on the feature maps processed according to the weighting, time for decoding may be reduced.

Abstract: Embodiments of this disclosure provide a video frame interpolation apparatus and method. The method includes calculating a bidirectional optical flow between a first frame and a second frame; an performing kernel and weight estimation according to the first frame and the second frame. An adaptive local convolutional kernel is generated by using a convolutional layer and a weight coefficient is generated by using another convolutional layer. A conversion on the first frame and the second frame is performed by using an adaptive conversion layer according to the bidirectional optical flow, the weight coefficient and the adaptive local convolutional kernel, so as to generate a conversion frame. A frame synthesis on the first frame, the second frame and the conversion frame is performed to generate an interpolation frame between the first frame and the second frame.

Abstract: Embodiments of this disclosure provide an image coding method and apparatus and an image decoding method and apparatus. The image coding method includes: performing feature extraction on to-be-processed image data by using a convolutional neural network, to generate feature maps of the image data; quantizing the feature maps to generate discrete feature maps; preprocessing the discrete feature maps to generate preprocessed data, an amount of data of the preprocessed data being less than an amount of data of the discrete feature maps; calculating probabilities of to-be-coded data in the discrete feature maps according to the preprocessed data; and performing entropy coding on the to-be-coded data according to the probabilities of the to-be-coded data.

Abstract: Embodiments of this disclosure provide a training method, an image encoding method, an image decoding method and apparatuses thereof. The image encoding apparatus includes: an image encoder configured to encode input image data to obtain a latent variable; a quantizer configured to perform quantizing processing on the latent variable according to a quantization step to generate a quantized latent variable; and an entropy encoder configured to perform entropy coding on the quantized latent variable by using an entropy model to form a bit stream.

Abstract: Embodiments of this disclosure provide an image coding apparatus, a probability model generating apparatus and an image decoding apparatus. A processor is to perform feature extraction on an input image to obtain first feature maps of N channels; to perform feature extraction on the input image with a size of the input image being adjusted K times, to respectively obtain second feature maps of N channels; and to concatenate the first feature maps of the K×N channels with the second feature maps of K×N channels to output a concatenated feature maps of channels. Hence, features of images may be accurately extracted and more competitive latent representations may be obtained.

Abstract: Embodiments of this disclosure provide an image coding method and apparatus and an image decoding method and apparatus. The image coding method includes: performing feature extraction on to-be-processed image data by using a convolutional neural network, to generate feature maps of the image data; quantizing the feature maps to generate discrete feature maps; preprocessing the discrete feature maps to generate preprocessed data, an amount of data of the preprocessed data being less than an amount of data of the discrete feature maps; calculating probabilities of to-be-coded data in the discrete feature maps according to the preprocessed data; and performing entropy coding on the to-be-coded data according to the probabilities of the to-be-coded data.

Abstract: Embodiments of this disclosure provide an image coding method and apparatus and an image compression system. The image coding apparatus includes a memory and a processor. The processor is configured to perform feature extraction on an input image to obtain feature maps of N channels; assign a weight to a feature map of each channel among the N channels; perform down-dimension processing on weighted feature maps processed in association with the N channels, to obtain feature maps of M channels and output the feature maps of M channels, M being smaller than N. Hence, by multiplying different feature maps by a weight to obtain corresponding importance and then performing down-dimension processing on the feature maps processed according to the weighting, time for decoding may be reduced.

Abstract: Embodiments of this disclosure provide an image encoding method and apparatus and image decoding method and apparatus. The image encoding includes performing convolutional neural network (CNN) encoding on image data to generate feature vectors or feature maps; quantizing the feature vectors or feature maps to generate discrete symbols to be encoded; and estimating probabilities of the symbols to be encoded by using a multi-scale context model including multiple mask convolution layers of different scales. An entropy encoding of the image data is performed according to the probabilities of the symbols to be encoded.