While deep neural networks (NN) significantly advance image compressed sensing (CS) by improving reconstruction quality, the necessity of training current CS NNs from scratch constrains their effectiveness and hampers rapid deployment. Although recent methods utilize pre-trained diffusion models for image reconstruction, they struggle with slow inference and restricted adaptability to CS. To tackle these challenges, this paper proposes Invertible Diffusion Models (IDM), a novel efficient, end-to-end diffusion-based CS method. IDM repurposes a large-scale diffusion sampling process as a reconstruction model, and fine-tunes it end-to-end to recover original images directly from CS measurements, moving beyond the traditional paradigm of one-step noise estimation learning. To enable such memory-intensive end-to-end fine-tuning, we propose a novel two-level invertible design to transform both (1) multi-step sampling process and (2) noise estimation U-Net in each step into invertible networks. As a result, most intermediate features are cleared during training to reduce up to 93.8% GPU memory. In addition, we develop a set of lightweight modules to inject measurements into noise estimator to further facilitate reconstruction. Experiments demonstrate that IDM outperforms existing state-of-the-art CS networks by up to 2.64dB in PSNR. Compared to the recent diffusion-based approach DDNM, our IDM achieves up to 10.09dB PSNR gain and 14.54 times faster inference. Code is available at https://github.com/Guaishou74851/IDM.
Download full-text PDF |
Source |
---|---|
http://dx.doi.org/10.1109/TPAMI.2025.3538896 | DOI Listing |
Cancer Cell
March 2025
Department of Pediatric Oncology, Dana-Farber/Boston Children's Cancer and Blood Disorder Center and Harvard Medical School, Boston, MA 02215, USA; Broad Institute of MIT and Harvard, Cambridge, MA 02142, USA. Electronic address:
PDGFRA is crucial to tumorigenesis and frequently genomically altered in high-grade glioma (HGG). In a comprehensive dataset of pediatric HGG (n = 261), we detect PDGFRA mutations and/or amplifications in 15% of cases, suggesting PDGFRA as a therapeutic target. We reveal that the PDGFRA/KIT inhibitor avapritinib shows (1) selectivity for PDGFRA inhibition, (2) distinct patterns of subcellular effects, (3) in vitro and in vivo activity in patient-derived HGG models, and (4) effective blood-brain barrier penetration in mice and humans.
View Article and Find Full Text PDFUltrason Sonochem
March 2025
College of Mechanical and Electrical Engineering, Inner Mongolia Agricultural University, Hohhot 010018, China.
To improve the efficiency of harvested quinoa seed wash processing, this study comparatively evaluated the effects of ultrasound-assisted hydration (UH) and conventional hydration (CH) on hydration dynamics, saponin mass transfer kinetics, and pericarp structural changes in quinoa seeds. Moisture uptake was monitored using a gravimetric method, saponin content was determined through enzyme-linked immunosorbent assay, and pericarp structural changes were observed via scanning electron microscopy. The results showed that UH significantly enhanced the water absorption rate of quinoa seeds, with the Peleg model effectively fitting all hydration stages (R > 0.
View Article and Find Full Text PDFMed Image Anal
February 2025
Wellcome Centre for Integrative Neuroimaging, FMRIB Centre, Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, United Kingdom; Department of Bioengineering, Imperial College London, London, United Kingdom.
Mapping how neurons are structurally wired into whole-brain networks can be challenging, particularly in larger brains where 3D microscopy is not available. Multi-modal datasets combining MRI and microscopy provide a solution, where high resolution but 2D microscopy can be complemented by whole-brain but lowresolution MRI. However, there lacks unified approaches to integrate and jointly analyse these multi-modal data in an insightful way.
View Article and Find Full Text PDFNeural Netw
March 2025
Department of Electrical Engineering, Indian Institute of Technology Madras (IITM), India; Healthcare Technology Innovation Centre, IITM, India.
Attention Mechanism (AM) selectively focuses on essential information for imaging tasks and captures relationships between regions from distant pixel neighborhoods to compute feature representations. Accelerated magnetic resonance image (MRI) reconstruction can benefit from AM, as the imaging process involves acquiring Fourier domain measurements that influence the image representation in a non-local manner. However, AM-based models are more adept at capturing low-frequency information and have limited capacity in constructing high-frequency representations, restricting the models to smooth reconstruction.
View Article and Find Full Text PDFPhys Rev Lett
February 2025
Institute of High Energy Physics, Key Laboratory of Particle Astrophysics & Experimental Physics Division & Computing Center, Chinese Academy of Sciences, 100049 Beijing, China.
The diffuse Galactic gamma-ray emission is a very important tool used to study the propagation and interaction of cosmic rays in the Milky Way. In this Letter, we report the measurements of the diffuse emission from the Galactic plane-covering Galactic longitudes from 15° to 235° and latitudes from -5° to +5°, in an energy range of 1 to 25 TeV-made with the Water Cherenkov Detector Array (WCDA) of the Large High Altitude Air Shower Observatory. After the sky regions of known sources are masked, the diffuse emission is detected with 24.
View Article and Find Full Text PDFEnter search terms and have AI summaries delivered each week - change queries or unsubscribe any time!