The paper "Feature Re-Embedding: Towards Foundation Model-Level Performance in Computational Pathology" addresses the limitations of existing multiple instance learning (MIL) frameworks in computational pathology, which often rely on offline instance feature extractors like pre-trained ResNet or foundation models. These extractors lack fine-tuning capabilities for specific downstream tasks, leading to suboptimal performance. To tackle this issue, the authors propose the Re-embedded Regional Transformer (R²T), a module that re-embeds instance features online, capturing fine-grained local features and establishing connections across different regions. Unlike previous works that focus on pre-training powerful feature extractors or designing sophisticated instance aggregators, R²T is tailored for online feature re-embedding and can be seamlessly integrated into mainstream MIL models. The key contributions of the paper include: 1. Introducing a novel paradigm for MIL models that incorporates a re-embedding module to improve the discriminative ability of instance features. 2. Designing R²T, which can be integrated into mainstream MIL models to enhance performance. The proposed R²T-MIL outperforms other methods by a significant margin on various computational pathology benchmarks. 3. Introducing two novel components: Cross-Region MSA (CR-MSA) and Embedded Position Encoding Generator (EPEG). CR-MSA enables effective information fusion across different regions, while EPEG combines relative and convolutional position encodings to encode positional information more effectively. Experimental results on common computational pathology tasks validate that: 1. Feature re-embedding improves the performance of MIL models based on ResNet-50 features to the level of foundation model features. 2. R²T introduces significant performance improvements to various MIL models. 3. R²T-MIL outperforms other latest methods by a large margin. The code for the proposed method is available at: <https://github.com/DearCaoR/RRT-MIL>.The paper "Feature Re-Embedding: Towards Foundation Model-Level Performance in Computational Pathology" addresses the limitations of existing multiple instance learning (MIL) frameworks in computational pathology, which often rely on offline instance feature extractors like pre-trained ResNet or foundation models. These extractors lack fine-tuning capabilities for specific downstream tasks, leading to suboptimal performance. To tackle this issue, the authors propose the Re-embedded Regional Transformer (R²T), a module that re-embeds instance features online, capturing fine-grained local features and establishing connections across different regions. Unlike previous works that focus on pre-training powerful feature extractors or designing sophisticated instance aggregators, R²T is tailored for online feature re-embedding and can be seamlessly integrated into mainstream MIL models. The key contributions of the paper include: 1. Introducing a novel paradigm for MIL models that incorporates a re-embedding module to improve the discriminative ability of instance features. 2. Designing R²T, which can be integrated into mainstream MIL models to enhance performance. The proposed R²T-MIL outperforms other methods by a significant margin on various computational pathology benchmarks. 3. Introducing two novel components: Cross-Region MSA (CR-MSA) and Embedded Position Encoding Generator (EPEG). CR-MSA enables effective information fusion across different regions, while EPEG combines relative and convolutional position encodings to encode positional information more effectively. Experimental results on common computational pathology tasks validate that: 1. Feature re-embedding improves the performance of MIL models based on ResNet-50 features to the level of foundation model features. 2. R²T introduces significant performance improvements to various MIL models. 3. R²T-MIL outperforms other latest methods by a large margin. The code for the proposed method is available at: <https://github.com/DearCaoR/RRT-MIL>.