Papers33

#vision-language models

Utonia: Toward One Encoder for All Point Clouds

Yujia Zhang, Xiaoyang Wu et al.Mar 3arXiv

Utonia is a single brain (encoder) that learns from many kinds of 3D point clouds, like indoor rooms, outdoor streets, tiny toys, and even city maps.

#Utonia#point cloud#self-supervised learning

AgilePruner: An Empirical Study of Attention and Diversity for Adaptive Visual Token Pruning in Large Vision-Language Models

Intermediate

Changwoo Baek, Jouwon Song et al.Mar 1arXiv

Big picture: Vision-language models look at hundreds of image pieces (tokens), which makes them slow and sometimes chatty with mistakes called hallucinations.

#visual token pruning#attention-based pruning#diversity-based pruning

Compositional Generalization Requires Linear, Orthogonal Representations in Vision Embedding Models

Intermediate

Arnas Uselis, Andrea Dittadi et al.Feb 27arXiv

The paper asks a simple question: what must a vision model’s internal pictures (embeddings) look like if it can recognize new mixes of things it already knows?

#compositional generalization#linear representation hypothesis#orthogonal representations

Retrieve and Segment: Are a Few Examples Enough to Bridge the Supervision Gap in Open-Vocabulary Segmentation?

Intermediate

Tilemachos Aravanis, Vladan Stojnić et al.Feb 26arXiv

This paper teaches an AI to segment any object you name (open-vocabulary) much better by adding a few example pictures with pixel labels and smart retrieval.

#open-vocabulary segmentation#vision-language models#retrieval-augmented

TOPReward: Token Probabilities as Hidden Zero-Shot Rewards for Robotics

Intermediate

Shirui Chen, Cole Harrison et al.Feb 22arXiv

Robots learn better when they get small hints at every step instead of only a final thumbs-up or thumbs-down.

#TOPReward#token probabilities#logits

AI Gamestore: Scalable, Open-Ended Evaluation of Machine General Intelligence with Human Games

Intermediate

Lance Ying, Ryan Truong et al.Feb 19arXiv

The paper argues that the fairest way to check how generally smart an AI is, is to see how quickly and well it learns lots of different human-made games, just like a person with the same time and practice.

#general intelligence#evaluation benchmark#game-based testing

ExStrucTiny: A Benchmark for Schema-Variable Structured Information Extraction from Document Images

Intermediate

Mathieu Sibue, Andres Muñoz Garza et al.Feb 12arXiv

ExStrucTiny is a new test (benchmark) that checks if AI can pull many connected facts from all kinds of documents and neatly put them into JSON, even when the question style and schema change.

#structured information extraction#document understanding#vision-language models

DeepImageSearch: Benchmarking Multimodal Agents for Context-Aware Image Retrieval in Visual Histories

Intermediate

Chenlong Deng, Mengjie Deng et al.Feb 11arXiv

Most image search systems judge each photo by itself, which fails when clues are split across many photos taken over time.

#context-aware image retrieval#multimodal agents#visual history exploration

Unveiling Implicit Advantage Symmetry: Why GRPO Struggles with Exploration and Difficulty Adaptation

Intermediate

Zhiqi Yu, Zhangquan Chen et al.Feb 5arXiv

The paper finds a hidden symmetry inside GRPO’s advantage calculation that accidentally stops models from exploring new good answers and from paying the right attention to easy versus hard problems at the right times.

#GRPO#GRAE#A-GRAE

Theory of Space: Can Foundation Models Construct Spatial Beliefs through Active Exploration?

Intermediate

Pingyue Zhang, Zihan Huang et al.Feb 4arXiv

This paper asks a simple question with big consequences: can today’s AI models actively explore a new space and build a trustworthy internal map of it?

#active exploration#cognitive map#spatial belief

SpatiaLab: Can Vision-Language Models Perform Spatial Reasoning in the Wild?

Intermediate

Azmine Toushik Wasi, Wahid Faisal et al.Feb 3arXiv

SpatiaLab is a new test that checks if vision-language models (VLMs) can understand real-world spatial puzzles, like what’s in front, behind, bigger, or reachable.

#SpatiaLab#spatial reasoning#vision-language models

Instruction Anchors: Dissecting the Causal Dynamics of Modality Arbitration

Intermediate

Yu Zhang, Mufan Xu et al.Feb 3arXiv

The paper asks a simple question: when an AI sees a picture and some text but the instructions say 'only trust the picture,' how does it decide which one to follow?

#multimodal instruction following#modality arbitration#instruction tokens

1 2 3