Local Texture Estimator for Implicit Representation Function

CVPR 2022  Β·  Jaewon Lee, Kyong Hwan Jin Β·

Recent works with an implicit neural function shed light on representing images in arbitrary resolution. However, a standalone multi-layer perceptron shows limited performance in learning high-frequency components. In this paper, we propose a Local Texture Estimator (LTE), a dominant-frequency estimator for natural images, enabling an implicit function to capture fine details while reconstructing images in a continuous manner. When jointly trained with a deep super-resolution (SR) architecture, LTE is capable of characterizing image textures in 2D Fourier space. We show that an LTE-based neural function achieves favorable performance against existing deep SR methods within an arbitrary-scale factor. Furthermore, we demonstrate that our implementation takes the shortest running time compared to previous works.

PDF Abstract CVPR 2022 PDF CVPR 2022 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Image Super-Resolution BSD100 - 2x upscaling LTE PSNR 32.44 # 4
Image Super-Resolution BSD100 - 3x upscaling LTE PSNR 29.39 # 3
Image Super-Resolution BSD100 - 4x upscaling LTE PSNR 27.86 # 3
Image Super-Resolution Set14 - 2x upscaling LTE PSNR 34.25 # 4
Image Super-Resolution Set14 - 3x upscaling LTE PSNR 30.8 # 2
Image Super-Resolution Set14 - 4x upscaling LTE PSNR 29.06 # 2
Image Super-Resolution Set5 - 2x upscaling LTE PSNR 38.33 # 3
Image Super-Resolution Set5 - 3x upscaling LTE PSNR 34.89 # 1
Image Super-Resolution Set5 - 4x upscaling LTE PSNR 32.81 # 2
Image Super-Resolution Urban100 - 2x upscaling LTE PSNR 33.5 # 4
Image Super-Resolution Urban100 - 3x upscaling LTE PSNR 29.41 # 2
Image Super-Resolution Urban100 - 4x upscaling LTE PSNR 27.24 # 3

Methods


No methods listed for this paper. Add relevant methods here