Frozen Giants, Tiny Tweaks: Lightweight Adaptation Boosts Fashion Image Search
Researchers in India show that training only a 2.4-million-parameter projection module on top of frozen CLIP encoders lifts fashion image-to-text ...
Tag: vision-language models
AI Learns When to Distrust Its Own Tools, Cutting Errors and Halving Costs
A new conflict-aware framework teaches vision-language models to align their own uncertainty with tool confidence, reject unreliable OCR output, and ...
Hypergraph Adapters Push Parameter-Efficient Multimodal Fine-Tuning to New Heights
Researchers have introduced HGA-Net, a hypergraph-based adapter that connects related examples within each mini-batch and achieves near-perfect fine-grained recognition with ...
From CLIP to Llama 4: A sweeping review maps the rise of pretrained multimodal AI
A new comprehensive review analyzes 13 pretrained multimodal AI models from 2020 to 2025, comparing their architectures, datasets, and applications ...
Frozen AI Models Learn to Watch Hours of Video Without Any Training
A new survey maps how training-free pipelines built on frozen AI models are tackling hour-long video understanding through token selection, ...
AI Without Glosses: How Large Language Models Are Learning to Translate Sign Language
A new survey maps how large language models and vision-language models are enabling gloss-free sign language translation to break communication ...
AI System Fact-Checks Charts by Finding the Right One Among Millions
Researchers at IIT Jodhpur have built a retrieval-augmented AI framework and a massive new dataset, FactViz3M, that can locate the ...
Teaching Machines to See and Say Defects: New AI Network Fuses Images with Language for Factory Inspections
A new cross-modal AI framework called CMAC-Net aligns visual defect regions with textual semantics, improving industrial anomaly detection accuracy and ...
New AI System Writes Eye Ultrasound Reports and Explains Them
Researchers have developed OphthUS-GPT, a multimodal AI system that automatically generates and interprets diagnostic reports from ophthalmic B-scan ultrasound images ...
AI Learns to Distrust Its Own Illusions: CLIP Helps Models Adapt Without Source Data
Researchers have developed SRRI, a source-free domain adaptation method that uses CLIP as an external verifier to de-bias pseudo-source domains ...
-
Mothers who receive childcare support from maternal grandparents show more parental warmth, finds NTU Singapore study
-
University of Seville Breaks 120-Year-Old Mystery, Revises a Key Einstein Concept
-
Bee body mass, pathogens and local climate influence heat tolerance
-
Researchers record first-ever images and data of a shark experiencing a boat strike
-
Groundbreaking Clinical Trial Reveals Lubiprostone Enhances Kidney Function
Embark on a thrilling journey of discovery with Scienmag.com—your ultimate source for cutting-edge breakthroughs. Immerse yourself in a world where curiosity knows no limits and tomorrow’s possibilities become today’s reality!