Text Alone Predicts Which Classes a Segmentation Readout Keeps
cs.CV, cs.AI
Submitted: 2026-11-09
Updated: 2026-11-09
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Understanding intermediate layers using linear classifier probes
- Conformal Risk Control
- The devil is in the fine-grained details: Evaluating open-vocabulary object detectors for fine-grained understanding
- What a MESS: Multi-Domain Evaluation of Zero-Shot Semantic Segmentation
- Compute-Optimal Pretrain--Fine-tune in Ridge Gradient Descent
- TIPSv2: Advancing Vision-Language Pretraining with Enhanced Patch-Text Alignment
- SAM 3: Segment Anything with Concepts
- An Empirical Study and Analysis of Generalized Zero-Shot Learning for Object Recognition in the Wild
- Probing Representation Forgetting in Supervised and Unsupervised Continual Learning
- Towards Continual Test-Time Adaptation of Vision-Language Models in Open-Vocabulary Semantic Segmentation
- On Calibration of Modern Neural Networks
- Pay Attention to Your Neighbours: Training-Free Open-Vocabulary Semantic Segmentation
- Active Testing: Sample-Efficient Model Evaluation
- Rotational Equilibrium: How Weight Decay Balances Learning Across Neural Networks
- Readout Blindness: VLM Scores Miss the Spatial Direction Their Frozen Encoders Retain
- Open-Vocabulary Semantic Segmentation with Mask-adapted CLIP
- Stepping Out of Similar Semantic Space for Open-Vocabulary Segmentation
- What CLIP Knows but Cannot Say: Recovering Negation from Frozen Intermediate Features
- Gradient Descent Maximizes the Margin of Homogeneous Neural Networks
- Long-tail learning via logit adjustment
Related papers
- TEA: Text Encoder Alignment for Robust Concept Erasure in Text-to-Image Models
- Do Egocentric Video-Language Models Capture Both Hand- and Object-Centric Cues?
- Aura: Consistent Multi-Subject Video Generation via VLM-Grounded Semantic Alignment
- PVRF: All-in-one Adverse Weather Removal via Prior-modulated and Velocity-constrained Rectified Flow
- DLWM: Diverse Latent World Models for Efficient Multimodal Reasoning
- Out of Sight, Still in Mind: Token Compression for Omni-LLMs