Tagged articles
visual language models
20 articles · Page 1 of 1
The paper reveals that visual language models often rely on chronological bias, mistaking later frames for progress, and introduces EgoTSR—a 46‑million‑sample ego‑centric dataset and three‑stage curriculum that teaches models to assess task state, evaluate with forward‑reverse tests, and achieve over 92% accuracy on long‑term robotic tasks.
Roboticschronological-biascurriculum learning
0 likes · 11 min read

Contra Labs' Design Crit dataset reveals that while AI can generate images, current AI judges barely outperform random guessing in assessing design quality, but a small fine‑tuned model can close nearly half the gap to human agreement by learning from expert‑rated criteria.
AI design evaluationDesign Crit datasetGenerative AI
0 likes · 16 min read

This paper builds a multi‑scale candlestick‑chart dataset and a standardized evaluation framework to measure how well visual language models (VLMs) extract price information, using confusion‑matrix diagnostics and Information Coefficient (IC) metrics, and finds that VLMs excel only on monotonic trends and struggle with precise time‑based predictions.
Prompt engineeringcandlestick chartinformation coefficient
0 likes · 13 min read

MIT and IBM researchers introduce ChartNet, the largest code‑guided synthetic chart dataset containing 1.5 million multimodal samples across 24 chart types and six libraries, and demonstrate that fine‑tuning visual‑language models on it yields consistent, significant gains on chart reconstruction, data extraction, summarization, and reasoning tasks, outperforming much larger off‑the‑shelf models including GPT‑4o.
AI researchChartNetMultimodal Dataset
0 likes · 13 min read

The paper presents RTPrune, a token‑pruning technique for DeepSeek‑OCR that exploits a two‑stage reading behavior in LLM decoding, first keeping high‑norm visual tokens and then fusing the rest via optimal‑transport matching with a dynamic pruning‑rate strategy, achieving up to 15% GFLOPs reduction and 18.9% speedup while preserving over 99% OCR accuracy across multiple benchmarks.
DeepSeek-OCROCR efficiencyOptimal Transport
0 likes · 9 min read

The paper introduces Visual Para-Thinker, a parallel‑thinking framework for large‑scale visual‑language models that uses visual‑centered block and scan path partitions, Path‑aware Attention and Learnable Parallel Rotary Position Embedding, and demonstrates consistent gains across counting, visual search, hallucination and grounding benchmarks.
LPRoPEPa-Attentionbenchmark evaluation
0 likes · 11 min read


SuanNiMar 27, 2026 · Artificial Intelligence
The OmniScience project introduces a 1.5‑million high‑quality image‑text pair dataset and a sophisticated pipeline that parses complex scientific documents, rewrites figure captions with large language models, and dramatically improves multimodal AI performance on benchmark tests.
Data Annotationmultimodal AIscientific dataset
0 likes · 9 min read

The article analyzes why standard prompt tuning (CoOp) causes catastrophic forgetting in visual‑language models, introduces the KgCoOp framework that adds a knowledge‑guided loss to regularize prompts, and shows through extensive experiments on 11 benchmarks that KgCoOp improves unseen‑class accuracy, harmonic mean, and efficiency while discussing trade‑offs and limitations.
Catastrophic ForgettingKnowledge-guided OptimizationPrompt Tuning
0 likes · 11 min read

The VLAW framework, developed by researchers from Tsinghua and Stanford, integrates high‑fidelity world models with visual‑language models, enabling real‑time physical interaction and intent understanding, which could dramatically improve training efficiency for embodied robots and mark a milestone toward safe, autonomous agents in complex real‑world environments.
Embodied AIRoboticsSimulation
0 likes · 6 min read

The article explains how contrastive learning replaces fixed‑category visual training with a relationship‑based approach, detailing the dual‑encoder architecture, cosine similarity loss, batch scaling, temperature control, zero‑shot capabilities, scalability from web data, and the method's strengths and limitations in modern multimodal AI.
CLIPContrastive LearningZero-Shot Learning
0 likes · 25 min read

This article presents a low‑resource offensive content detection framework that leverages multi‑agent visual‑language models (MA‑VLMs) for self‑training and a novel Positive‑Negative‑Unlabeled (PNU) loss, enabling accurate classification with as few as 50 annotated samples across multimodal datasets.
Content moderationPNU lossSelf‑Training
0 likes · 20 min read


Amap TechJul 14, 2025 · Artificial Intelligence
The UPRE framework introduces multi‑view domain prompts and unified representation enhancement to achieve zero‑shot domain adaptation for object detection, dramatically improving detection accuracy on unseen target domains across diverse visual scenarios.
Object DetectionPrompt engineeringcross-domain learning
0 likes · 10 min read

This article systematically introduces the most common contrastive learning loss functions—including Contrastive Loss, Triplet Loss, N‑pair Loss, InfoNCE, and Cross‑Entropy—explaining their mathematical formulations, advantages, challenges, and typical applications in visual, textual, and multimodal representation learning.
Contrastive LearningInfoNCELoss Functions
0 likes · 10 min read

This article explains contrastive learning for visual language models, covering its definition, four‑step workflow, how to choose positive and negative pairs, the difference between supervised and self‑supervised variants, and why the technique is essential for zero‑shot and cross‑modal capabilities.
Contrastive LearningData AugmentationRepresentation Learning
0 likes · 6 min read

This article introduces visual‑language models (VLMs), explaining how they combine large language models with visual encoders, why they overcome the rigidity of traditional computer‑vision systems, their key advantages, modular architecture, training methods, and practical applications such as image captioning and visual question answering.
AI applicationsComputer VisionContrastive Learning
0 likes · 8 min read


AIWalkerMay 26, 2025 · Artificial Intelligence
VisionReasoner presents a reinforcement‑learning‑driven unified framework that simultaneously tackles detection, segmentation, and counting tasks, employing a novel multi‑target cognition strategy and efficient Hungarian‑based matching, and demonstrates substantial gains—29.1% on COCO detection, 22.1% on ReasonSeg, and 15.3% on CountBench—using only 7,000 training samples.
Object DetectionReinforcement LearningVisionReasoner
0 likes · 20 min read

SuperEdit introduces rectified instruction generation and contrastive supervision to fix noisy supervision in instruction‑based image editing, achieving up to 9.19% performance gains on Real‑Edit benchmarks without extra model parameters or pre‑training, and releases all data and code publicly.
diffusion modelsimage editingvisual language models
0 likes · 15 min read

The article analyzes how reinforcement learning, which transformed LLM reasoning in DeepSeek, can be applied to visual‑language models to overcome the limitations of traditional chain‑of‑thought prompting and supervised fine‑tuning, presenting concrete reward designs, training pipelines, and a critical assessment of their strengths and weaknesses.
LLMRL trainingReinforcement Learning
0 likes · 10 min read

MiniGPT-5 is a novel multimodal generation model using generative vokens to interleave text and image synthesis, integrating Stable Diffusion and LLMs with a two-stage training that requires no domain-specific annotations, achieving state‑of‑the‑art coherence and quality on benchmarks like CC3M, VIST, and MMDialog.
AI researchMultimodal GenerationStable Diffusion
0 likes · 9 min read

Recent advances in visual language models enable zero-shot multimodal tasks, and this article explores their application to open-set object detection, prompt learning, and promptable surgical instrument segmentation, highlighting methods like CLIP, CoOp, and the DetPro framework with experimental results across multiple benchmarks.
Computer Visionmultimodalopen-set detection
0 likes · 12 min read
