ConvNeXt
PulseAugur coverage of ConvNeXt — every cluster mentioning ConvNeXt across labs, papers, and developer communities, ranked by signal.
1 day(s) with sentiment data
-
6G Robotic Vehicle Networks Enhanced by Knowledge Distillation and GANs
Researchers have developed a novel framework called KDG-SemNOMA for 6G robotic vehicle networks, aiming to improve visual perception under bandwidth and energy constraints. This framework utilizes a ConvNeXt-based DeepJ…
-
New CGS method enhances large-kernel CNNs for mobile deployment
Researchers have introduced a new technique called Channel Group-Shared (CGS) low-rank approximation to make large-kernel Convolutional Neural Networks (CNNs) more efficient for deployment on mobile devices. This method…
-
Frozen foundation model embeddings enable label-efficient underwater image classification
Researchers have demonstrated a label-efficient method for underwater image classification using frozen foundation model embeddings. By employing DINOv3 ViT-B/16 embeddings and training only a logistic regression classi…
-
New multimodal learning improves fracture classification accuracy
Researchers have developed a multimodal learning approach to improve the accuracy of fracture classification from radiographs, particularly when patient metadata is incomplete or mismatched. Their method combines a Conv…
-
New pruning method significantly reduces Vision Transformer computational costs
Researchers have developed a new method called Denoised Variance-Based Pruning with Optimal Brain Bias Compensation (DVBP + OB$^2$C) to reduce the computational overhead of Vision Transformers (ViTs). This technique use…
-
New XtraLight-MedMamba model classifies precancerous polyps with high accuracy
Researchers have developed XtraLight-MedMamba, a novel deep learning framework designed for classifying neoplastic tubular adenomas from whole-slide images. This architecture combines a ConvNeXt feature extractor with v…
-
New AQUA20 dataset targets challenging underwater species classification
Researchers have introduced AQUA20, a new benchmark dataset designed to improve underwater species classification. The dataset contains 8,171 images of 20 marine species, specifically curated to address challenges like …
-
FFNet introduces efficient convolutional mixer for vision tasks
Researchers have introduced FFNet, a novel convolutional mixer architecture designed for enhanced efficiency in computer vision tasks. FFNet reinterprets the Feed-Forward Network (FFN) component of Transformers as a mem…
-
New benchmark compares visual backbones for solar irradiance forecasting
A new research paper introduces a controlled benchmark for evaluating visual backbones in multimodal short-term solar irradiance forecasting. The study fixes the overall forecasting pipeline and isolates the visual back…
-
Harmonized ECG features improve cross-dataset clinical prediction
Researchers have developed a harmonized and interpretable feature representation for electrocardiogram (ECG) waveforms to improve cross-dataset clinical prediction. This approach, called FeatureDB, aims to reduce perfor…
-
Pretraining domain crucial for private medical image AI, study finds
A new study published on arXiv investigates the impact of pretraining data on the privacy-utility trade-off in medical image analysis models. Researchers found that the domain of the pretraining data, specifically using…
-
Foundation models make multi-branch fusion less effective for vehicle re-ID
A new research paper questions the effectiveness of multi-branch architectures and the fusion of different backbone models for vehicle re-identification tasks in the era of foundation models. The study found that a sing…
-
New hierarchical ensemble methods improve zebrafish phenotype classification
Researchers have developed and evaluated three hierarchical ensemble methods for classifying zebrafish phenotypes from embryo images. The study compared three backbone architectures: ResNet18, ViT, and ConvNeXt. ConvNeX…
-
AI systems advance ambivalence and hesitancy recognition in video analysis · 8 sources tracked
Researchers have developed advanced methods for recognizing ambivalence and hesitancy in videos, participating in the 11th ABAW Challenge. One approach, the HSEmotion team's system, utilizes multi-task learning with fro…
-
New SEAMS method identifies crucial image regions for AI model behavior
Researchers have developed SEAMS, a novel saliency method designed to identify image regions crucial for preserving a model's behavior. This approach optimizes a soft mask using a preservation objective, directly search…
-
AI system EcoVision uses drone imagery for salt marsh vegetation mapping
Researchers have developed EcoVision, an AI-powered system utilizing drone imagery for monitoring salt marsh vegetation. The system employs transformer-based semantic segmentation and a ConvNeXt architecture for fine-gr…
-
New LFNet method fuses CNN and SSM features for improved salient object detection
Researchers have developed a novel method called Liquid Fusion Network (LFNet) to improve salient object detection by harmonizing features from different neural network architectures. LFNet addresses the spectral biases…
-
Deep learning models improve breathing cessation detection in preterm infants
Researchers have developed deep learning models to more accurately detect cessation of breathing events in preterm infants within neonatal intensive care units. These models, utilizing impedance pneumography (IP), elect…
-
AI Transfer Attacks: "Scissors Effect" Reveals Diversity Hinders Robust Models
Researchers have identified a phenomenon called the "Scissors Effect" in transfer attacks against AI models. This effect demonstrates that while random resizing and padding (Input Diversity or DI) generally improve atta…
-
New capsule architecture enhances gaze estimation accuracy and speed
Researchers have developed CapStARE, a novel capsule-based architecture for gaze estimation. This system utilizes a frozen ConvNeXt backbone for efficient feature extraction and capsule formation with attention-based ro…