# US 2026/0311292 A1 · Cascaded local implicit transformer for arbitrary-scale super-resolution

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0311292 A1
- Document ID: US20260311292A1
- Status: Published application
- Kind: A1
- Application number: 19/697,598
- Filed: 2026-06-04
- Published: 2026-10-08
- Earliest claimed: 2022-08-26
- Assignees: Mediatek Inc.; National Tsing Hua University
- Applicants: Mediatek Inc.; National Tsing Hua University
- Inventors: Yu-Syuan Xu; Hao-Wei Chen; Chun-Yi Lee
- Web page: https://patentgrep.com/patent/US20260311292A1

## CPC

- G06T 3/4046: using neural networks
- G06T 3/4053: based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
- G06T 7/73: using feature-based methods
- G06T 2207/20081: Training; Learning
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

A method can include receiving a low-resolution (LR) image, extracting a first feature embedding from the LR image, performing a first upsampling to the LR image by a first upsampling factor to generate a upsampled image, receiving a LR coordinate of a pixel within the LR image and a first cell size of the LR coordinate, generating a first residual image based on the first feature embedding, the LR coordinate, and the first cell size of the LR coordinate using a local implicit image function, and generating a first high-resolution (HR) image by combining the first residual image and the upsampled image via element-wise addition.

## Claims

### Claim 1

A method, comprising:
receiving a low-resolution (LR) image;
extracting a first feature embedding from the LR image;
performing a first upsampling to the LR image by a first upsampling factor to generate an upsampled image;
receiving a LR coordinate of a pixel within the LR image and a first cell size of the LR coordinate;
generating a first residual image based on the first feature embedding, the LR coordinate, and the first cell size of the LR coordinate using a local implicit image function; and
generating a first high-resolution (HR) image by combining the first residual image and the upsampled image;
wherein the first residual image is generated by:
applying convolution functions to the first feature embedding to extract a local query latent embedding, a local key latent embedding and a local value latent embedding at the LR coordinate;
performing upsampling operations to the local query latent embedding, and sampling the local key latent embedding and the local value latent embedding; and
generating a local latent embedding.

### Claim 2 (depends on 1)

The method of claim 1, wherein the local latent embedding is generated by:
generating an attention matrix by adding an inner product of the local query latent embedding and the local key latent embedding with a relative positional bias of the LR coordinate;
generating a local attention map by normalizing the attention matrix; and
generating the local latent embedding by performing element-wise multiplication of the local value latent embedding and the local attention map.

### Claim 3 (depends on 2)

The method of claim 2, wherein the relative positional bias of the LR coordinate is generated by processing a set of local relative coordinates through a positional encoding block and a fully-connected layer.

### Claim 4 (depends on 1)

The method of claim 1, wherein performing the upsampling operations to the local query latent embedding, and sampling the local key latent embedding and the local value latent embedding comprises:
performing a first sampling operation on the local query latent embedding according to a target high-resolution (HR) coordinate; and
performing a second sampling operation on the local key latent embedding and the local value latent embedding according to a plurality of low-resolution (LR) coordinates within a local grid.

### Claim 5 (depends on 4)

The method of claim 4, wherein the first sampling operation comprises performing a bilinear interpolation on the local query latent embedding, and wherein the second sampling operation comprises sampling the local key latent embedding and the local value latent embedding at the plurality of LR coordinates within the local grid.

### Claim 6 (depends on 4)

The method of claim 4, wherein the local grid is positioned at an LR coordinate that is closest to the target HR coordinate.

### Claim 7 (depends on 1)

The method of claim 1, wherein applying the convolution functions to the first feature embedding comprises:
projecting the first feature embedding into a query feature map, a key feature map, and a value feature map using a plurality of separate convolutional layers,
wherein the local query latent embedding, the local key latent embedding, and the local value latent embedding are extracted respectively from the query feature map, the key feature map, and the value feature map.

### Claim 8

An apparatus, comprising circuitry configured to:
receive a low-resolution (LR) image;
extract a first feature embedding from the LR image;
perform a first upsampling to the LR image by a first upsampling factor to generate an upsampled image;
receive a LR coordinate of a pixel within the LR image and a first cell size of the LR coordinate;
generate a first residual image based on the first feature embedding, the LR coordinate, and the first cell size of the LR coordinate by using a local implicit image function; and
generate a first high-resolution (HR) image by combining the first residual image and the upsampled image;
wherein the first residual image is generated by:
applying convolution functions to the first feature embedding to extract a local query latent embedding, a local key latent embedding and a local value latent embedding at the LR coordinate;
performing upsampling operations to the local query latent embedding, and sampling the local key latent embedding and the local value latent embedding; and
generating a local latent embedding.

### Claim 9 (depends on 8)

The apparatus of claim 8, wherein the local latent embedding is generated by:
generating an attention matrix by adding an inner product of the local query latent embedding and the local key latent embedding with a relative positional bias of the LR coordinate;
generating a local attention map by normalizing the attention matrix; and
generating the local latent embedding by performing element-wise multiplication of the local value latent embedding and the local attention map.

### Claim 10 (depends on 9)

The apparatus of claim 9, wherein the relative positional bias of the LR coordinate is generated by processing a set of local relative coordinates through a positional encoding block and a fully-connected layer.

### Claim 11 (depends on 8)

The apparatus of claim 8, wherein the apparatus is further configured to:
perform a first sampling operation on the local query latent embedding according to a target high-resolution (HR) coordinate; and
perform a second sampling operation on the local key latent embedding and the local value latent embedding according to a plurality of low-resolution (LR) coordinates within a local grid.

### Claim 12 (depends on 11)

The apparatus of claim 11, wherein the first sampling operation comprises performing a bilinear interpolation on the local query latent embedding, and wherein the second sampling operation comprises sampling the local key latent embedding and the local value latent embedding at the plurality of LR coordinates within the local grid.

### Claim 13 (depends on 11)

The apparatus of claim 11, wherein the local grid is positioned at an LR coordinate that is closest to the target HR coordinate.

### Claim 14 (depends on 8)

The apparatus of claim 8, the apparatus is further configured to:
project the first feature embedding into a query feature map, a key feature map, and a value feature map using a plurality of separate convolutional layers,
wherein the local query latent embedding, the local key latent embedding, and the local value latent embedding are extracted respectively from the query feature map, the key feature map, and the value feature map.
