Abstract:Light-weight super-resolution (SR) models have received considerable attention for their serviceability in mobile devices. Many efforts employ network quantization to compress SR models. However, these methods suffer from severe performance degradation when quantizing the SR models to ultra-low precision (e.g., 2-bit and 3-bit) with the low-cost layer-wise quantizer. In this paper, we identify that the performance drop comes from the contradiction between the layer-wise symmetric quantizer and the highly asymmetric activation distribution in SR models. This discrepancy leads to either a waste on the quantization levels or detail loss in reconstructed images. Therefore, we propose a novel activation quantizer, referred to as Dynamic Dual Trainable Bounds (DDTB), to accommodate the asymmetry of the activations. Specifically, DDTB innovates in: 1) A layer-wise quantizer with trainable upper and lower bounds to tackle the highly asymmetric activations. 2) A dynamic gate controller to adaptively adjust the upper and lower bounds at runtime to overcome the drastically varying activation ranges over different <a class="link-external link-http" href="http://samples.To" rel="external noopener nofollow">this http URL</a> reduce the extra overhead, the dynamic gate controller is quantized to 2-bit and applied to only part of the SR networks according to the introduced dynamic intensity. Extensive experiments demonstrate that our DDTB exhibits significant performance improvements in ultra-low precision. For example, our DDTB achieves a 0.70dB PSNR increase on Urban100 benchmark when quantizing EDSR to 2-bit and scaling up output images to x4. Code is at \url{<a class="link-external link-https" href="https://github.com/zysxmu/DDTB" rel="external noopener nofollow">this https URL</a>}.

Super-Resolving Barcode Images with an Edge-Preserving Variational Bayesian Framework

Robust Dual Images Super-Resolution.

Multi-frame Image Super Resolution with Natural Image Prior

Robust Super Resolution of Compressed Video

Boosting Image Super-Resolution Via Fusion of Complementary Information Captured by Multi-Modal Sensors

Root exudates in relation to growth and nitrogenase activity of Rhizobium japonicum.

Bayesian Image Super-Resolution with Deep Modeling of Image Statistics.

Single Remote Sensing Image Super-Resolution Via a Generative Adversarial Network with Stratified Dense Sampling and Chain Training

QR code resolution improvement based on Super-Resolution Generative Adversarial Network

Image Super-resolution Reconstruction Algorithm Based on Bayesian Theory

Efficient Barcode Localization Method for Low-Quality Images

A Hardware Realization of Superresolution Combining Random Coding and Blurring

Unified Regularization Framework for Blind Image Super-Resolution

Image Super-Resolution Using Multi-Layer Support Vector Regression

Blind Image Super-Resolution: A Survey and Beyond

A Fast Super-Resolution Method Based on Sparsity Properties.

Spatially-Variant Degradation Model for Dataset-free Super-resolution

Identification and Recovery of the Blurred QR Code Image

A Multi-Frame Image Super-Resolution Method.

Dynamic Dual Trainable Bounds for Ultra-low Precision Super-Resolution Networks

Breaking the Limitations with Sparse Inputs by Variational Frameworks (BLIss) in Terahertz Super-Resolution 3D Reconstruction