Hybrid Mamba-Transformer Vision UNet for Efficient Medical Image Segmentation
A novel hybrid model, HMT-UNet, that combines the strengths of State Space Models (SSMs) and Transformers to achieve efficient and effective medical image segmentation.
Advancing Zero-shot Polyp Segmentation in Colorectal Cancer Detection with Segment Anything Model 2
The Segment Anything Model 2 (SAM 2) demonstrates superior performance in zero-shot polyp segmentation tasks compared to its predecessor SAM and other state-of-the-art methods, facilitating efficient and accurate image and video polyp segmentation for improved clinical workflows and early cancer detection.
고해상도 범분광 영상을 활용한 가중치 기반 대표 계수 총 변동 정규화를 통한 고해상도 초분광 영상 잡음 제거
고해상도 범분광 영상을 활용하여 가중치 기반 대표 계수 총 변동 정규화 기법을 통해 고해상도 초분광 영상의 잡음을 효과적으로 제거할 수 있다.
Enhancing Hyperspectral Image Quality through Panchromatic-Guided Denoising
A novel pan-denoising framework that leverages panchromatic (PAN) images to guide the denoising of hyperspectral images (HSIs), achieving superior performance in removing various types of noise while preserving important spatial and spectral details.
포괄적인 조직병리학 기반 모델을 이용한 난소암 아형 분류 평가
조직병리학 기반 모델은 난소암 아형 분류 성능을 크게 향상시킬 수 있다.
Comprehensive Evaluation of Histopathology Foundation Models for Ovarian Cancer Subtype Classification
Histopathology foundation models significantly outperform ImageNet-pretrained models in classifying ovarian cancer subtypes, with the largest models achieving over 90% balanced accuracy.
Enhancing Robustness of Under-Display Camera Image Restoration Models Against Adversarial Attacks
This study proposes a novel defense framework that combines adversarial purification and fine-tuning techniques to enhance the robustness of under-display camera image restoration models against various adversarial attacks.
Automated Segmentation of Diffuse Large B-Cell Lymphoma Lesions in PET/CT Images Using Multi-Scale Information Fusion and Cross-Attention
This study presents an effective automatic segmentation method for diffuse large B-cell lymphoma (DLBCL) that leverages the complementary strengths of PET and CT imaging through multi-scale information fusion and cross-attention mechanisms.
High-resolution Compressing for OCR-free Multi-page Document Understanding
A novel layout-aware compressing architecture, High-resolution DocCompressor, is proposed to greatly reduce the number of visual tokens for high-resolution document images while maintaining most visual information. This enables state-of-the-art performance on multi-page document understanding benchmarks with significantly faster inference speed.
Guide-and-Rescale: A Self-Guidance Mechanism for Effective and Tuning-Free Real Image Editing
The proposed Guide-and-Rescale method leverages a self-guidance mechanism to effectively manipulate real images while preserving the original image structure and details.
