<ul><li>Existing dataset pruning techniques primarily focus on classification tasks, limiting their applicability to more complex and practical tasks like instance segmentation.</li><li>A novel Training-Free Dataset Pruning (TFDP) method is proposed for instance segmentation, addressing the challenges of pixel-level annotations, instance area variations, and class imbalances.</li><li>The method leverages shape and class information from image annotations to design a Shape Complexity Score (SCS), refining it into Scale-Invariant (SI-SCS) and Class-Balanced (CB-SCS) versions.</li><li>The proposed method achieves state-of-the-art results on VOC 2012, Cityscapes, and COCO datasets, and accelerates the pruning process by an average of 1349 times on COCO compared to adapted baselines.</li></ul>

Training-Free Dataset Pruning for Instance Segmentation

Discover more