LaRS: A Diverse Panoptic Maritime Obstacle Detection Dataset and Benchmark

Open paper
Benchmark Model Rank Results
panoptic-segmentation-on-larsMask2Former (Swin-B)#1PQ: 41.7
panoptic-segmentation-on-larsPanoptic FPN (ResNet-50)#2PQ: 40.1
panoptic-segmentation-on-larsMask2Former (Swin-T)#3PQ: 39.2
panoptic-segmentation-on-larsPanoptic FPN (ResNet-101)#4PQ: 38.7
panoptic-segmentation-on-larsMask2Former (ResNet-50)#5PQ: 37.6
panoptic-segmentation-on-larsMask2Former (ResNet-101)#6PQ: 37.2
panoptic-segmentation-on-larsPanoptic Deeplab (ResNet-50)#7PQ: 34.7
panoptic-segmentation-on-larsMaX-DeepLab#8PQ: 31.9
semantic-segmentation-on-larsKNet (Swin-T)#1Q: 71.3F1: 73.4μ: 78.8mIoU: 97.2
semantic-segmentation-on-larsSegFormer (MiT-B2)#2Q: 67.8F1: 70.0μ: 78.6mIoU: 96.8
semantic-segmentation-on-larsDeepLabv3 (ResNet-101)#3Q: 62.9F1: 66.1μ: 77.5mIoU: 95.2
semantic-segmentation-on-larsPointRend#4Q: 62.1F1: 65.4μ: 77.5mIoU: 94.9
semantic-segmentation-on-larsDeepLabv3+ (ResNet-101)#5Q: 61.0F1: 64.0μ: 77.8mIoU: 95.4
semantic-segmentation-on-larsSTDC2#6Q: 60.8F1: 64.3μ: 76.5mIoU: 94.5
semantic-segmentation-on-larsFCN (ResNet-101)#7Q: 60.2F1: 63.4μ: 77.4mIoU: 95.0
semantic-segmentation-on-larsWaSR (ResNet-101)#8Q: 59.5F1: 61.6μ: 71.0mIoU: 96.6
semantic-segmentation-on-larsSTDC1#9Q: 57.8F1: 61.8μ: 75.6mIoU: 93.6
semantic-segmentation-on-larsFCN (ResNet-50)#10Q: 53.6F1: 57.9μ: 76.8mIoU: 92.6
semantic-segmentation-on-larsSegmenter (ViT-B)#11Q: 52.6F1: 55.2μ: 72.2mIoU: 95.1
semantic-segmentation-on-larsBiSeNetv2#12Q: 51.2F1: 54.7μ: 73.9mIoU: 93.5
semantic-segmentation-on-larsWODIS (ResNet-101)#13Q: 40.7F1: 47.5μ: 63.0mIoU: 85.7
semantic-segmentation-on-larsBiSeNetv1 (ResNet-50)#14Q: 39.4F1: 42.8μ: 73.3mIoU: 92.2
semantic-segmentation-on-larsIntCatchAI#15Q: 20.5F1: 44.9μ: 62.4mIoU: 45.6
semantic-segmentation-on-larsUNet#16Q: 13.9F1: 15.4μ: 75.7mIoU: 90.1
video-semantic-segmentation-on-larsWaSR-T (ResNet-101)#1Q: 60.1F1: 62.1μ: 71.1mIoU: 96.7
video-semantic-segmentation-on-larsTMANet (ResNet-50)#2Q: 57.5F1: 61.1μ: 77.1mIoU: 94.1
video-semantic-segmentation-on-larsCSANet (ResNet-101)#3Q: 49.1F1: 52.1μ: 63.7mIoU: 94.2

Rank counts only results with a code link.