# US 2026/0087337 A1 · Neural processing unit and method for multi-stage video quality improvement > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2026/0087337 A1 - Document ID: US20260087337A1 - Status: Published application - Kind: A1 - Application number: 19/347,769 - Filed: 2025-10-02 - Published: 2026-03-26 - Earliest claimed: 2021-04-01 - Assignees: Deepx Co., Ltd. - Applicants: Deepx Co., Ltd. - Inventors: Lok Won KIM - Web page: https://patentgrep.com/patent/US20260087337A1 ## CPC - G06N 3/063: using electronic means - G06N 3/045: Combinations of networks - G06N 3/0455: Auto-encoder networks; Encoder-decoder networks - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/0495: Quantised networks; Sparse networks; Compressed networks - G06N 3/09: Supervised learning - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract A neural processing unit (NPU) includes an internal memory storing information on combinations of a plurality of artificial neural network (ANN) models, the plurality of ANN models including first and second ANN models; a plurality of processing elements (PEs) to process first operations and second operations of the plurality of ANN models in sequence or in parallel, the plurality of PEs including first and second groups of PEs; and a scheduler to allocate to the first group of PEs a part of the first operations for the first ANN model and to allocate to the second group of PEs a part of the second operations for the second ANN model, based on an instruction related to information on an operation sequence of the plurality of ANN models or further based on ANN data locality information. The first and second operations may be performed in parallel or in a time division. ## Claims ### Claim 1 A method for improving video quality using a neural processing unit (NPU), the method comprising: receiving, by the NPU, a video stream; decompressing or decoding, by a first artificial neural network (ANN) model executed on the NPU, the received video stream; performing, by a second ANN model executed on the NPU, a video preprocessing on the decompressed or decoded video stream; and applying, by a third ANN model executed on the NPU, a super-resolution process to the preprocessed video stream to generate an enhanced video stream. ### Claim 2 (depends on 1) The method of claim 1, wherein the video preprocessing comprises at least one of deblurring, denoising, applying a wide dynamic range (WDR) or high dynamic range (HDR) process, color tone mapping, or demosaicing. ### Claim 3 (depends on 1) The method of claim 1, wherein the first, second, and third ANN models are executed on different groups of processing elements (PEs) within the NPU. ### Claim 4 (depends on 3) The method of claim 3, wherein the groups of PEs are allocated by a scheduler within the NPU. ### Claim 5 (depends on 4) The method of claim 4, wherein the allocation is based on at least one of an operation amount or a priority associated with each of the first, second, and third ANN models. ### Claim 6 (depends on 1) The method of claim 1, wherein the decompressing or decoding, the performing the video preprocessing, and the applying the super-resolution process are performed sequentially in a pipeline. ### Claim 7 (depends on 1) The method of claim 1, wherein the NPU is part of an edge device. ### Claim 8 (depends on 7) The method of claim 7, wherein the edge device is a camera device. ### Claim 9 (depends on 1) The method of claim 1, wherein the super-resolution process increases the resolution of the video stream. ### Claim 10 An edge device comprising: a camera for capturing a video stream; and a neural processing unit (NPU) configured to: receive the captured video stream; decompress or decode the received video stream using a first artificial neural network (ANN) model; perform a video preprocessing on the decompressed or decoded video stream using a second ANN model; and apply a super-resolution process to the preprocessed video stream to generate an enhanced video stream using a third ANN model. ### Claim 11 (depends on 10) The edge device of claim 10, wherein the video preprocessing comprises: at least one of deblurring, denoising, applying a wide dynamic range (WDR) or high dynamic range (HDR) process, color tone mapping, or demosaicing. ### Claim 12 (depends on 10) The edge device of claim 10, wherein the NPU comprises: a plurality of processing elements (PEs), and wherein the first, second, and third ANN models are executed on different groups of the PEs. ### Claim 13 (depends on 12) The edge device of claim 12, further comprising: a scheduler configured to allocate the groups of PEs. ### Claim 14 (depends on 13) The edge device of claim 13, wherein the scheduler is configured to: allocate the groups of PEs based on at least one of an operation amount or a priority associated with each of the first, second, and third ANN models. ### Claim 15 (depends on 10) The edge device of claim 10, further comprising: a display for displaying the enhanced video stream. ### Claim 16 (depends on 10) The edge device of claim 10, further comprising: a memory configured to store the enhanced video stream. ### Claim 17 A neural processing unit (NPU) comprising: a plurality of processing elements (PEs); and a scheduler configured to: allocate a first group of PEs to execute a first artificial neural network (ANN) model for decompressing or decoding a video stream; allocate a second group of PEs to execute a second ANN model for performing video preprocessing on the decompressed or decoded video stream; and allocate a third group of PEs to execute a third ANN model for applying a super-resolution process to the preprocessed video stream to generate an enhanced video stream. ### Claim 18 (depends on 17) The NPU of claim 17, wherein the video preprocessing comprises: at least one of deblurring, denoising, applying a wide dynamic range (WDR) or high dynamic range (HDR) process, color tone mapping, or demosaicing. ### Claim 19 (depends on 17) The NPU of claim 17, wherein the first, second, and third groups of PEs are at least partially different. ### Claim 20 (depends on 17) The NPU of claim 17, wherein the scheduler is further configured to: perform the allocations based on at least one of an operation amount or a priority associated with each of the first, second, and third ANN models.