Image Classification

Wasserstein Distance-based Knowledge Distillation (WKD) rivals KL-divergence by leveraging rich category interrelations and handling non-overlapping distributions, significantly boosting performance i…

VMamba: a vision backbone achieving linear time complexity using Visual State Space (VSS) blocks and 2D Selective Scan (SS2D) for efficient visual representation.

Visual pinwheel centers in the cortex act as efficient geometric saliency detectors, responding faster and stronger to complex spatial textures than other structures.

Visual Fourier Prompt Tuning (VFPT) leverages the Fast Fourier Transform to seamlessly integrate spatial and frequency information for superior parameter-efficient vision model fine-tuning, even with …

CONBIAS tackles dataset bias by representing visual data as concept graphs, diagnosing imbalances via clique analysis, and debiasing through targeted data augmentation for improved model generalizatio…

OoD-ViT-NAS: a new benchmark reveals how ViT architecture impacts out-of-distribution generalization, highlighting the importance of embedding dimension and challenging the reliance on in-distribution…

Deep learning models favor simple features, hindering generalization; this paper introduces a new feature complexity metric revealing a spectrum of simple-to-complex features, their learning dynamics,…

Researchers unveil a novel framework to dissect bias in large-scale visual datasets, identifying unique visual attributes and leveraging language models for detailed analysis, paving the way for creat…

Transformer Doctor diagnoses and treats vision transformer errors by identifying and correcting information integration issues, improving model performance and interpretability.

TrAct boosts vision model training by directly optimizing first-layer activations, leading to significant speedups (1.25x-4x) and improved accuracy.

This paper introduces novel metrics for visual emotion recognition evaluation, considering the psychological distance between emotions to better reflect human perception, improving the assessment of m…

Using real images retrieved from a generator’s training data outperforms using synthetic images generated by that same model for image classification.

TARP-VP reveals a surprising lack of trade-off between adversarial robustness and privacy for label mapping visual prompting models, showing that transferred adversarial training significantly improve…

Structured Unrestricted-Rank Matrices (SURMs) revolutionize parameter-efficient fine-tuning by offering greater flexibility and accuracy than existing methods like LoRA, achieving significant gains in…

Spiking Experts Mixture Mechanism (SEMM) boosts Spiking Transformers by integrating Mixture-of-Experts for efficient, sparse conditional computation, achieving significant performance improvements on …

Scala: One-shot training enables flexible ViT inference!

CRATE-a: A new white-box vision transformer architecture achieves 85.1% ImageNet accuracy by strategically scaling model size and datasets, outperforming prior white-box models and preserving interpre…

ScaleKD: Pre-trained vision transformers make excellent teachers for diverse student networks, improving efficiency and performance in knowledge distillation.

Samba: a novel severity-aware recurrent model, tackles cross-domain medical image grading by sequentially encoding image patches and recalibrating states using EM, significantly improving accuracy.

Boosting AI’s continual learning via saliency-driven experience replay, achieving up to 20% accuracy improvement.