Sirvi Autor "Prytula, Yaroslav" järgi

Nüüd näidatakse 1 - 1 1

IAUNet: Instance-Aware U-Net
(Tartu Ülikool, 2025) Prytula, Yaroslav; Fishman, Dmytro, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
Instance segmentation is critical in biomedical imaging to accurately distinguish individual objects like cells, which often overlap and vary in size. Recent query-based methods, where object queries guide segmentation, have shown strong performance. While U-Net has been a go-to architecture for medical image segmentation, its potential in query-based approaches remains largely unexplored. In this work, we present IAUNet, a novel query-based U-Net architecture. The core design keeps the full U-Net structure and introduces a lightweight convolutional Pixel decoder that efficiently aggregates multi-scale features with minimal computational cost, making the model more efficient and reducing the number of parameters unlike its Transformer counterparts. On top of that, we propose a Transformer decoder with deep supervision that refines object-specific queries across multiple layers and resolutions, enabling precise instance-level segmentation. Finally, we introduce the Revvity-25 Full Cell Segmentation Dataset, a new 2025 benchmark featuring detailed annotations of overlapping cell cytoplasm in brightfield images. This dataset provides high-resolution labels with accurate instance boundaries and supports evaluation under both modal and amodal segmentation settings. Extensive experiments on multiple public datasets and Revvity-25, we show that IAUNet outperforms most state-of-the-art fully convolutional, transformer-based, and query-based models and cell segmentation-specific models, establishing a strong baseline for cell instance segmentation tasks.