lpips
PulseAugur coverage of lpips — every cluster mentioning lpips across labs, papers, and developer communities, ranked by signal.
9 day(s) with sentiment data
-
Xiaomi MiLM Plus releases PROVE benchmark for video object removal
Xiaomi's MiLM Plus has introduced PROVE, a new benchmark and set of metrics designed to evaluate video object removal models more effectively. Traditional metrics like PSNR and SSIM struggle with the inherently ill-pose…
-
New framework CodecArena assesses video codec quality using visual reinforcement learning
Researchers have introduced CodecArena, a novel vision-language framework designed to assess video codec quality, particularly in low and ultra-low bitrate scenarios. Unlike existing metrics that focus on feature simila…
-
New method tackles ambiguous matches in video frame interpolation
Researchers have developed a novel framework for video frame interpolation that addresses the challenge of ambiguous matches in optical flow estimation. This new method preserves multiple candidate correspondences and u…
-
New framework reconstructs video from event streams using appearance engrams
Researchers have developed Engram-E2VID, a novel framework for reconstructing target RGB frames from an event stream and a reference frame. This method addresses the challenge of associating event-derived structures wit…
-
New diffusion models enhance pathological image resolution for better diagnostics · 2 sources tracked
Two new research papers propose advanced diffusion models for enhancing the resolution of pathological images, aiming to improve diagnostic accuracy. S$^3$-Diff utilizes a Structural Semantic Synergy approach with speci…
-
New diffusion model generates satellite imagery using wavelet-domain conditioning
Researchers have developed a novel diffusion framework that utilizes wavelet-domain conditioning for generating satellite imagery from cartographic data. This approach, built upon ControlNet and a Stable Diffusion backb…
-
New WorldDynCache framework speeds up diffusion world models
Researchers have developed WorldDynCache, a novel framework designed to accelerate the inference process for diffusion world models. This system employs a risk-controlled latent dynamics approximation to mitigate the co…
-
New LAGS method uses graph learning for efficient 3D drone scene reconstruction
Researchers have developed LAGS, a new method for 3D scene reconstruction using aerial drone imagery. To address inefficiencies in resource allocation, they propose a groupwise heterogeneous graph neural network (GW-HGN…
-
New framework enables reference-free evaluation of camera ISP pipelines
Researchers have developed a novel reference-free framework for evaluating image signal processing (ISP) pipelines in cameras. This method estimates full-reference image quality metrics like PSNR, SSIM, and LPIPS by pre…
-
New dual-stream learning enhances electron microscopy imaging
Researchers have developed a novel frequency-aware dual-stream learning architecture to improve electron microscopy imaging. This approach decomposes images into low-frequency structures and high-frequency details, usin…
-
MIRAGE model enhances MRI contrast enhancement prediction
Researchers have developed MIRAGE, a novel 2D U-Net model designed to infer contrast enhancement in breast MRIs from pre-contrast slices. The model integrates global reconstruction and perceptual losses with specialized…
-
New framework uses VLMs to improve EEG-to-image reconstruction evaluation
Researchers have developed a new framework to evaluate the coherence between EEG signals and reconstructed images, addressing limitations in existing metrics like SSIM and LPIPS. This framework utilizes four Vision-Lang…
-
New framework analyzes CT reconstruction, noise impacts strategy
Researchers have developed a unified framework to analyze design choices in self-supervised sparse-view CT reconstruction. Their experiments on simulated and real-world datasets indicate that the optimal partitioning st…
-
SalientGS unifies SfM and 3DGS for faster 3D scene reconstruction · 2 sources tracked
Researchers have developed SalientGS, a novel pipeline that unifies Structure-from-Motion (SfM) with 3D Gaussian Splatting (3DGS) for 3D scene reconstruction. The system employs importance-guided Markov Chain Monte Carl…
-
Simon-SR framework enhances image super-resolution with prompt-guided adaptation
Researchers have introduced Simon-SR, a novel multi-modal framework designed to enhance single-image super-resolution (SISR) by leveraging learnable prompts for semantic mining and text-image fusion. This approach aims …
-
Diffusion model enhances face recognition from low-quality surveillance images
Researchers have developed FASR++, a new diffusion model designed to improve face recognition accuracy from low-quality surveillance images. This model aggregates features from multiple low-resolution images to generate…
-
New DTI paradigm enhances generative face video super-resolution
Researchers have introduced a new paradigm called Dynamic Trajectory Initialization (DTI) for Generative Face Video Super-Resolution (GFVSR). This method reformulates GFVSR as an input-driven directional restoration pro…
-
ReScene framework reconstructs 3D indoor scenes with improved accuracy · arXiv paper
Researchers have developed ReScene, a new framework designed to construct simulation-ready 3D indoor scenes from multi-view captures. This method addresses limitations in existing approaches by focusing on cross-view re…
-
AI data quality metrics misaligned with human perception and task performance
A new paper published on arXiv explores the disconnect between automated data quality metrics and their actual utility for deep learning models, particularly in Earth observation. The research highlights that common met…
-
Study finds common AI data quality metrics unreliable for Earth observation
A new study has revealed that common data-quality metrics used for evaluating synthetic datasets in deep learning are unreliable, particularly for Earth observation data. Metrics like Fréchet Inception Distance (FID) an…