📢 Check out our work on Snuffy: Efficient Whole Slide Image Classifier 💉🔬, where we addressed computational efficiency, accuracy, and robustness in processing Whole Slide Pathology Images, presented at #ECCV2024! 🔍 Abstract: WSI classification faces computational and performance challenges. We introduce Snuffy, a novel MIL-pooling method using a novel sparse transformer to improve performance with minimal pre-training. Snuffy allows for continual few-shot pre-training and provides the best probabilistic bound on sparse transformer layers for universal approximation. Results on CAMELYON16 and TCGA Lung cancer datasets show superior performance at both WSI and patch levels. 🔗 Project: [jafarinia.github.io/snuffy_p…](jafarinia.github.io/snuffy_p…) 📄 Paper: [arxiv.org/abs/2408.08258](arxiv.org/abs/2408.08258) 💻 Code: [github.com/jafarinia/snuffy](github.com/jafarinia/snuffy) 🤗 Huggingface: [huggingface.co/nialda/snuffy](huggingface.co/nialda/snuffy) ⚡️ TL;DR: Snuffy reduces pre-training needs for WSI classification and offers state-of-the-art accuracy using a sparse transformer and the code actually works. 1/2 @eccvconf #WSI #DeepLearning #Transformers #SparseTransformers #DigitalPathology #SSL #MIL