# Image Super-Resolution

7 related articles

Subscribe to updates

arXiv Nov 10, 2026
CURRENT fills in both space and time for aortic 4D flow MRI with one recurrent network, sharpening relative pressure estimates
A team at the University of Zurich and ETH Zurich built CURRENT, a recurrent neural network that performs spatial, temporal and spatiotemporal super-resolution of aortic 4D flow MRI within one architecture. On synthetic data its spatial super-resolution directional error was 0.072±0.037 versus 0.097±0.053 for 4DFlowNet, and its temporal directional error was 0.062±0.031 versus 0.064±0.031 for T4DFlowNet. Joint spatiotemporal super-resolution gave relative pressure estimates closer to the high-resolution reference than either direction alone, and the model, trained only on synthetic data, showed promising generalization to in vivo acquisitions.
A team at the University of Zurich and ETH Zurich built CURRENT, a recurrent neural network that performs spatial, temporal and spatiotemporal super-resolution of aortic 4D flow MRI within one architecture. On synthetic data its spatial super-resolution directional error was 0.072±0.037 versus 0.097±0.053 for 4DFlowNet, and its temporal directional error was 0.062±0.031 versus 0.064±0.031 for T4DFlowNet. Joint spatiotemporal super-resolution gave relative pressure estimates closer to the high-resolution reference than either direction alone, and the model, trained only on synthetic data, showed promising generalization to in vivo acquisitions.
ETH Zurich
Luuk Jacobs（First author, Corresponding author）
CURRENT 4D Flow MRI Image Super-Resolution Recurrent Neural Networks
bioRxiv Oct 20, 2026
Flowing cells get organelle-level eyes: three structured-light frames plus self-supervised reconstruction reach about 118 nm at up to roughly 720 cells per second
Super-resolution imaging flow cytometry (SR-IFC) brings structured-illumination microscopy into a continuous microfluidic stream: each cell yields only three phase-shifted frames, physical modulation sharpens the flow axis, and a network trained solely on the imaged sample's own data infers the missing transverse detail. On 100 nm fluorescent beads, SR-IFC reached about 118 nm along the flow axis and 110 nm across it, versus 222 to 227 nm for wide-field imaging of the same beads; with a 2048 × 320 pixel readout, the theoretical throughput ceiling is about 720 cells per second.
Super-resolution imaging flow cytometry (SR-IFC) brings structured-illumination microscopy into a continuous microfluidic stream: each cell yields only three phase-shifted frames, physical modulation sharpens the flow axis, and a network trained solely on the imaged sample's own data infers the missing transverse detail. On 100 nm fluorescent beads, SR-IFC reached about 118 nm along the flow axis and 110 nm across it, versus 222 to 227 nm for wide-field imaging of the same beads; with a 2048 × 320 pixel readout, the theoretical throughput ceiling is about 720 cells per second.
Okayama University Hospital Tohoku University Okayama University Wuhan University University of California, Los Angeles Ludwig-Maximilians-Universität München The University of Tokyo Kindai University
Yingdong Luo（First author）Keisuke Goda（Corresponding author）Tianben Ding（Corresponding author）
SR-IFC Structured Illumination Microscopy Image Super-Resolution Self-Supervised Learning
arXiv Oct 24, 2026
Brain connectomics stops scanning every slice slowly: ZoomSEM rebuilds 16× detail from real low-resolution SEM scans
Serial electron microscopy is the workhorse of brain connectomics, yet imaging a cubic millimeter of brain takes months to years. ZoomSEM trains a super-resolution model on real low- and high-resolution SEM image pairs and recovers 4×, 8× and up to 16× higher-resolution images from fast low-resolution scans while preserving the fine neuronal structures connectomic analysis needs, reaching up to 30-fold effective acceleration. Experiments across datasets of diverse biological samples show robust generalization under realistic low-resolution acquisition.
Serial electron microscopy is the workhorse of brain connectomics, yet imaging a cubic millimeter of brain takes months to years. ZoomSEM trains a super-resolution model on real low- and high-resolution SEM image pairs and recovers 4×, 8× and up to 16× higher-resolution images from fast low-resolution scans while preserving the fine neuronal structures connectomic analysis needs, reaching up to 30-fold effective acceleration. Experiments across datasets of diverse biological samples show robust generalization under realistic low-resolution acquisition.
Fudan University Southeast University
Jiateng Shou（First author）
ZoomSEM Image Super-Resolution serial electron microscopy connectomics
researchsquare.com Oct 19, 2026
X-ray super-resolution stops borrowing a natural-image eye: swapping VGG perceptual loss lifts ESRGAN from 22.21 to 34.04 dB PSNR
Medical X-ray super-resolution has long relied on VGG perceptual loss designed for natural images, and reconstructed structure can drift from anatomy. A domain-aware loss called USELoss replaces VGG features with multi-scale features from a lightweight U-Net encoder and adds a Sobel edge constraint, applied with identical loss weights to both ESRGAN and HAT. On the PadChest chest X-ray test set, ESRGAN rises from 22.21 to 34.04 dB PSNR and from 0.5545 to 0.9160 SSIM, with edge error cut by about 46%.
Medical X-ray super-resolution has long relied on VGG perceptual loss designed for natural images, and reconstructed structure can drift from anatomy. A domain-aware loss called USELoss replaces VGG features with multi-scale features from a lightweight U-Net encoder and adds a Sobel edge constraint, applied with identical loss weights to both ESRGAN and HAT. On the PadChest chest X-ray test set, ESRGAN rises from 22.21 to 34.04 dB PSNR and from 0.5545 to 0.9160 SSIM, with edge error cut by about 46%.
Guru Gobind Singh Indraprastha University Lala Lajpat Rai Memorial Medical College
Aryan Tyagi（First author）
USELoss (U-Net Sobel Edge Loss)Image Super-Resolution U-Net
arXiv Oct 15, 2026
Vascular CT super-resolution stops chasing pixels: vessels get their own 3D label space, and broken portal veins reconnect
Vessels reconstructed from low-resolution contrast-enhanced CT can come out broken, spurious, or wrongly connected, and image-quality metrics do not capture those structural errors. VascuSR separates vascular structure from CT intensity reconstruction: a conditional 3D rectified-flow refiner corrects vessels in label space, a lightweight controller injects the refined structure into the diffusion decoder, and each new super-resolved volume re-extracts the structure for the next round. On re-annotated MSD Task08 and HVM, it beats interpolation and learning-based baselines in Dice and clDice, cuts class-averaged connected-component counts by over 20% on both datasets and mean centerline distance by over 10% on MSD, while keeping image quality comparable to its SR backbone.
Vessels reconstructed from low-resolution contrast-enhanced CT can come out broken, spurious, or wrongly connected, and image-quality metrics do not capture those structural errors. VascuSR separates vascular structure from CT intensity reconstruction: a conditional 3D rectified-flow refiner corrects vessels in label space, a lightweight controller injects the refined structure into the diffusion decoder, and each new super-resolved volume re-extracts the structure for the next round. On re-annotated MSD Task08 and HVM, it beats interpolation and learning-based baselines in Dice and clDice, cuts class-averaged connected-component counts by over 20% on both datasets and mean centerline distance by over 10% on MSD, while keeping image quality comparable to its SR backbone.
University of Michigan University of California San Diego Massachusetts Institute of Technology
Hairong Ni（First author）
VascuSR Image Super-Resolution Rectified Flow CT Reconstruction Diffusion Model
researchsquare.com Oct 14, 2026
Conglomerate digital rock stops choosing: wide-field scans recover pore detail, permeability error falls to 18.56%
To resolve the field-of-view versus resolution trade-off in micro-CT imaging of conglomerate reservoirs, Hy-MambaIR chains windowed self-attention with a Mamba state-space model into a local-to-global reconstruction network that recovers high-resolution pore and gravel structure from low-resolution wide-field slices. On paired multi-resolution micro-CT data it reaches a peak signal-to-noise ratio of 33.36 dB, 1.31 dB above HAT, and hits 30 dB with 6.6-fold fewer training iterations; on an 855³-voxel representative volume the mean relative error of three-directional absolute permeability is 18.56%, falling to 3.59% after supervised fine-tuning on pebbly sandstone.
To resolve the field-of-view versus resolution trade-off in micro-CT imaging of conglomerate reservoirs, Hy-MambaIR chains windowed self-attention with a Mamba state-space model into a local-to-global reconstruction network that recovers high-resolution pore and gravel structure from low-resolution wide-field slices. On paired multi-resolution micro-CT data it reaches a peak signal-to-noise ratio of 33.36 dB, 1.31 dB above HAT, and hits 30 dB with 6.6-fold fewer training iterations; on an 855³-voxel representative volume the mean relative error of three-directional absolute permeability is 18.56%, falling to 3.59% after supervised fine-tuning on pebbly sandstone.
China University of Petroleum, Beijing
Yumin Li（First author）Yang Zhang（Corresponding author）
Hy-MambaIR Image Super-Resolution Digital Rock Physics Mamba Transformer
arXiv Oct 8, 2026
On a 640-KB-SRAM STM32H563ZI MCU, an INT8-quantized super-resolution network simultaneously reconstructs high-resolution depth and intensity from a VL53L9CX SPAD sensor
This work presents a two-branch super-resolution deep-learning framework that simultaneously upscales low-resolution depth and intensity images from a consumer-grade VL53L9CX SPAD time-of-flight sensor, using a two-stage PixelShuffle reconstruction head and simplified sensor firmware to control activation memory, then exports the model to ONNX, quantizes it to INT8, generates C code, and deploys it on a bare-metal STM32H563ZI MCU with only 640 KB SRAM and 2 MB Flash for fully on-device super-resolution inference.
This work presents a two-branch super-resolution deep-learning framework that simultaneously upscales low-resolution depth and intensity images from a consumer-grade VL53L9CX SPAD time-of-flight sensor, using a two-stage PixelShuffle reconstruction head and simplified sensor firmware to control activation memory, then exports the model to ONNX, quantizes it to INT8, generates C code, and deploys it on a bare-metal STM32H563ZI MCU with only 640 KB SRAM and 2 MB Flash for fully on-device super-resolution inference.
Zhenya Zang（First author）
Image Super-Resolution VL53L9CX SPAD ToF Sensor Time-of-Flight Imaging Int8 Quantization Embedded Deep Learning

You’re all caught up

CURRENT fills in both space and time for aortic 4D flow MRI with one recurrent network, sharpening relative pressure estimates

Flowing cells get organelle-level eyes: three structured-light frames plus self-supervised reconstruction reach about 118 nm at up to roughly 720 cells per second

Brain connectomics stops scanning every slice slowly: ZoomSEM rebuilds 16× detail from real low-resolution SEM scans

X-ray super-resolution stops borrowing a natural-image eye: swapping VGG perceptual loss lifts ESRGAN from 22.21 to 34.04 dB PSNR

Vascular CT super-resolution stops chasing pixels: vessels get their own 3D label space, and broken portal veins reconnect

Conglomerate digital rock stops choosing: wide-field scans recover pore detail, permeability error falls to 18.56%

On a 640-KB-SRAM STM32H563ZI MCU, an INT8-quantized super-resolution network simultaneously reconstructs high-resolution depth and intensity from a VL53L9CX SPAD sensor