Computer Vision

DEX boosts CNN accuracy on tiny AI accelerators by 3.5%p, utilizing unused memory and processors to extend input channels without increasing latency.

DeTrack revolutionizes visual object tracking with an in-model latent denoising learning process, achieving real-time speed and state-of-the-art accuracy.

Depth Anywhere enhances 360-degree monocular depth estimation by cleverly using perspective models to label unlabeled 360-degree data, significantly improving accuracy.

Depth Anything V2 drastically improves monocular depth estimation by using synthetic training data, scaling up the teacher model, and employing pseudo-labeled real images. It outperforms previous met…

DEPrune: A novel GPU-optimized pruning method for depthwise separable convolutions, achieving up to 3.74x speedup on EfficientNet-B0 with no accuracy loss!

Vision’s Mamba model demystified: Researchers unveil its surprising link to linear attention, improving efficiency and accuracy through design enhancements.

DeltaDEQ accelerates deep equilibrium model inference by 73-84% via a novel ‘heterogeneous convergence’ exploitation technique, maintaining accuracy.

DEL: Learns 3D particle dynamics from 2D images via physics-informed neural rendering, exceeding existing methods’ accuracy and robustness.

AdvUnlearn enhances diffusion model robustness against adversarial attacks during concept erasure by integrating adversarial training, improving the trade-off between robustness and model utility.

Researchers developed semantic RSMs, a novel approach to measure semantic similarity in neural networks, improving image retrieval and aligning network representations with predicted class probabiliti…

Improved Kullback-Leibler (IKL) divergence loss achieves state-of-the-art adversarial robustness and competitive knowledge distillation performance by addressing KL loss’s limitations.

This paper presents a general framework for interpreting Vision Transformer (ViT) components, mapping their contributions to CLIP space for textual interpretation, and introduces a scoring function fo…

DeBaRA: a novel denoising-based model generates realistic & controllable 3D room layouts, surpassing existing methods.

AI-generated images contaminate online continual learning datasets, hindering performance. A new method, ESRM, leverages entropy and real/synthetic similarity maximization to select high-quality data…

Deep Degradation Response (DDR) uses image deep feature changes under degradation to create a flexible image descriptor, excelling in blind image quality assessment and unsupervised image restoration.

DCDepth achieves state-of-the-art monocular depth estimation by progressively predicting depth in the frequency domain via DCT, capturing local correlations and global context effectively.

DC-Gaussian: A novel method generates high-fidelity novel views from dashcam videos by addressing common windshield obstructions (reflections, occlusions) using adaptive image decomposition, illumina…

Unlearning synthesized images efficiently reveals influential training data for text-to-image models, improving data attribution accuracy and facilitating better model understanding.

DarkSAM, a novel prompt-free attack, renders the Segment Anything Model incapable of segmenting objects across diverse images, highlighting its vulnerability to universal adversarial perturbations.

DA-Ada enhances domain adaptive object detection by using a novel domain-aware adapter that leverages both domain-invariant and domain-specific knowledge for improved accuracy and generalization acros…