SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers

Benchmark Model Rank Results
2d-semantic-segmentation-on-wildscenesSegformer (MiT-B5)#5mIoU: 40.83
crack-segmentation-on-crackvision12kSegFormer#4mIoU: 0.57969
semantic-segmentation-on-ade20kSegFormer-B5#89Validation mIoU: 51.8Params (M): 84.7
semantic-segmentation-on-ade20kSegFormer-B4#97Validation mIoU: 51.1Params (M): 64.1
semantic-segmentation-on-ade20kSegFormer-B0#215Validation mIoU: 37.4Params (M): 3.8
semantic-segmentation-on-ade20k-valSegFormer-B5(MS, 87M #Params, ImageNet-1K pretrain)#43mIoU: 51.8
semantic-segmentation-on-cityscapesSegFormer (MiT-B5, Mapillary)#17Mean IoU (class): 83.1%
semantic-segmentation-on-cityscapes-valSegFormer (MiT-B5, Mapillary)#19mIoU: 84.0
semantic-segmentation-on-cityscapes-valSegFormer-B0#93Validation mIoU: 76.2
semantic-segmentation-on-dada-segSegFormer (MiT-B3)#10mIoU: 27.0
semantic-segmentation-on-dada-segSegFormer (MiT-B2)#18mIoU: 21.2
semantic-segmentation-on-dada-segSegFormer (MiT-B1)#26mIoU: 16.6
semantic-segmentation-on-ddd17SegFormer-B2#5mIoU: 71.05
semantic-segmentation-on-deliver-1SegFormer#9mIoU: 57.20
semantic-segmentation-on-densepassSegFormer (MiT-B2)#12mIoU: 42.4%
semantic-segmentation-on-densepassSegFormer (MiT-B1)#16mIoU: 38.5%
semantic-segmentation-on-dsecSegFormer-B2#4mIoU: 71.99
semantic-segmentation-on-eventscapeSegFormer-B4#3mIoU: 59.86
semantic-segmentation-on-eventscapeSegFormer-B2#4mIoU: 58.69
semantic-segmentation-on-fine-grained-grassSegFormer#4mIoU: 48.29
semantic-segmentation-on-potsdamSegFormer-B2#3mIoU: 84.65
semantic-segmentation-on-potsdamSegFormer-B1#4mIoU: 84.37
semantic-segmentation-on-potsdamSegFormer-B0#7mIoU: 83.67
semantic-segmentation-on-selmaSegFormer#2mIoU: 77.2
semantic-segmentation-on-spectralwasteSegFormer (HYPER)#3mIoU: 54.3
semantic-segmentation-on-spectralwasteSegFormer (HYPER3)#4mIoU: 53.5
semantic-segmentation-on-spectralwasteSegFormer (RGB)#5mIoU: 48.4
semantic-segmentation-on-synpassSegFomrer#3mIoU: 37.24%
semantic-segmentation-on-synthetic-bathingSegFormer#4mIoU: 86.86
semantic-segmentation-on-uplightSegFormer-B2 (RGB)#4mIoU: 89.60
semantic-segmentation-on-urbanlfSegFormer#8mIoU (Syn): 78.53mIoU (Real): 82.20
semantic-segmentation-on-us3dSegFormer-B2#4mIoU: 75.14
semantic-segmentation-on-us3dSegFormer-B1#6mIoU: 74.19
semantic-segmentation-on-us3dSegFormer-B0#9mIoU: 71.80
semantic-segmentation-on-vaihingenSegFormer-B1#5mIoU: 76.92
semantic-segmentation-on-vaihingenSegFormer-B2#8mIoU: 76.69
semantic-segmentation-on-vaihingenSegFormer-B0#10mIoU: 75.57
semantic-segmentation-on-zju-rgb-pSegFormer-B2 (RGB)#7mIoU: 89.6
thermal-image-segmentation-on-mfn-datasetSegFormer (B4)#26mIOU: 54.8
thermal-image-segmentation-on-mfn-datasetSegFormer (B2)#30mIOU: 53.2
thermal-image-segmentation-on-rgb-t-glassSegFormer#9MAE: 0.053