REVIEW 23 cited by
Vision language models are blind: Failing to translate detailed visual features into words
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Vision language models are blind: Failing to translate detailed visual features into words
read the original abstract
While large language models with vision capabilities (VLMs), e.g., GPT-4o and Gemini 1.5 Pro, score high on many vision-understanding benchmarks, they are still struggling with low-level vision tasks that are easy to humans. Specifically, on BlindTest, our suite of 7 very simple tasks, including identifying (a) whether two circles overlap; (b) how many times two lines intersect; (c) which letter is being circled in a word; and (d) the number of circles in an Olympic-like logo, four state-of-the-art VLMs are only 58.07% accurate on average. Claude 3.5 Sonnet performs the best at 77.84% accuracy, far from the human expected accuracy of 100%. Across different image resolutions and line widths, VLMs including slow-thinking models consistently struggle with those tasks that require precise spatial information when geometric primitives overlap or are close. Yet, VLMs perform at near-100% accuracy when much more space is added to separate shapes and letters. Linear probing experiments show that vision encoders contain sufficient visual information to solve BlindTest and that language models fail to decode this information into correct answers. Code and data are at: https://vlmsareblind.github.io
Forward citations
Cited by 23 Pith papers
-
The Inattentional Gap: Task-Conditioned Language and Vision Models Omit the Safety-Critical Signals They Can Otherwise Report
Task-conditioning suppresses AI report of co-present safety-critical signals the same models report when unconstrained, decoupling benchmark safety from deployment safety.
-
Do vision-language models search like humans? Reasoning tokens as a reaction-time analog in classic visual-search paradigms
Vision-language models show some human-like patterns in visual search effort (flat for features, rising for conjunctions) but diverge on target-present vs absent slopes and enumeration accuracy when reasoning tokens p...
-
Beyond 'One Language, One Script': Quantifying Orthographic Bias in Multilingual VLMs with PuMVR
PuMVR benchmark shows VLMs exhibit script-dependent bias on Punjabi tasks with accuracy gaps up to 16% and script consistency rates as low as 24.8%, even when visual input is provided.
-
Zone of Proximal Policy Optimization: Teacher in Prompts, Not Gradients
ZPPO improves distillation to small vision-language models by using binary and negative candidate prompts plus a replay buffer for hard questions, outperforming standard distillation and GRPO on a 31-benchmark suite w...
-
SpatialAct: Probing Spatial Reasoning-to-Action Capabilities of VLM Agents in 3D Scenes
SpatialAct benchmark shows VLMs handle isolated spatial reasoning but fail to maintain coherent spatial beliefs and produce reliable actions in multi-turn 3D interactions, underperforming humans.
-
Vision Language Models Cannot Reason About Physical Transformation
Current VLMs cannot maintain transformation-invariant representations of number, length, volume or size and instead rely on textual invariance priors that reverse on matched non-conserving controls.
-
The Inattentional Gap: Task-Conditioned Language and Vision Models Omit the Safety-Critical Signals They Can Otherwise Report
Task conditioning suppresses safety-critical signal reporting in language and vision models that unconstrained versions report at higher rates, creating an inattentional gap that decouples benchmark safety from real-w...
-
The Last Visible Pixel: Probing Fine-Scale Perception in Vision-Language Models
FineSightBench reveals VLMs perceive patterns down to 12px but show persistent failures in fine-scale reasoning such as numeracy and sequencing.
-
MOSS-Video-Preview: Toward Real-Time Video Understanding via Cross-Attention
MOSS-Video-Preview introduces a cross-attention architecture and synthesized real-time QA data to enable continuous perception, answer revision, and faster inference in video-language models compared to decoder-only designs.
-
Decomposed On-Policy Distillation for Vision-Language Reasoning: Steering Gradients for Visual Grounding
Decomposes VLM distillation loss into orthogonal language and visual components and introduces Visual Gradient Steering to prioritize visual grounding over standard monolithic optimization.
-
Binding Visual Features Point by Point
Training VLMs to point via text induces serial processing that eliminates binding errors and enables compositional generalization on multi-object tasks.
-
Do Vision-Language Models Understand 3D Scenes or Just Catalogue Objects?
VLMs achieve 53-97% on volumetric rearrangement planning but only 6-45% on occlusion and under 7% on reflections in a new 3,034-sample benchmark, with white-box analysis localizing the failure to visual-token merger i...
-
Do Vision-Language Models Understand 3D Scenes or Just Catalogue Objects?
VLMs achieve 53-97% on rearrangement planning but only 6-45% on occlusion and under 7% on reflections, with failures localized to visual token compression after the vision encoder.
-
Mirror, Mirror on the Wall: Can VLM Agents Tell Who They Are at All?
Stronger VLM agents use mirror reflections for self-identification in controlled 3D tests, while weaker ones inspect but fail to extract or correctly attribute self-relevant information.
-
S2H-DPO: Hardness-Aware Preference Optimization for Vision-Language Models
S2H-DPO generates hierarchical prompt-driven preference pairs to improve multi-image reasoning in VLMs while keeping single-image performance intact.
-
ReflectCAP: Detailed Image Captioning with Reflective Memory
ReflectCAP distills model-specific hallucination and oversight patterns into Structured Reflection Notes that steer LVLMs toward more factual and complete image captions, reaching the Pareto frontier on factuality-cov...
-
MiMo-Embodied: X-Embodied Foundation Model Technical Report
MiMo-Embodied is a single foundation model that achieves state-of-the-art results on 17 embodied AI benchmarks and 12 autonomous driving benchmarks through multi-stage learning, curated data, and CoT/RL fine-tuning th...
-
How Well Does GPT-4o Understand Vision? Evaluating Multimodal Foundation Models on Standard Computer Vision Tasks
Multimodal foundation models achieve respectable but sub-specialist performance on semantic vision tasks and weaker results on geometric tasks when evaluated through prompt chaining on established benchmarks.
-
Grounded Reinforcement Learning for Visual Reasoning
ViGoRL introduces visually grounded RL that anchors reasoning steps to image coordinates and uses multi-turn zooming to outperform standard RL and supervised baselines on spatial and GUI reasoning benchmarks.
-
MathFlow: Enhancing the Perceptual Flow of MLLMs for Visual Mathematical Problems
MathFlow decouples perception and inference stages in MLLMs for visual math, with a dedicated perception model delivering gains on the FlowVerse benchmark when paired with existing reasoners.
-
Context Unrolling in Omni Models
Omni is a multimodal model whose native training on diverse data types enables context unrolling, allowing explicit reasoning across modalities to better approximate shared knowledge and improve downstream performance.
-
Information-Regularized Attention for Visual-Centric Reasoning
IRA is a stochastic attention mechanism that regulates visual information injection in VLMs to yield smoother embedding trajectories and reduced attention sinks.
-
Seed1.5-VL Technical Report
Seed1.5-VL is a compact multimodal model that sets new records on dozens of vision-language benchmarks and outperforms prior systems on agent-style tasks.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.