XCiT: Cross-Covariance Image Transformers

Benchmark Model Rank Results
image-classification-on-imagenetXCiT-L24#173Top 1 Accuracy: 86%Number of params: 189MGFLOPs: 417.9
image-classification-on-imagenetXCiT-M24#187Top 1 Accuracy: 85.8%Number of params: 84MGFLOPs: 188
image-classification-on-imagenetXCiT-S24#208Top 1 Accuracy: 85.6%Number of params: 48MGFLOPs: 106
image-classification-on-imagenetXCiT-S12#252Top 1 Accuracy: 85.1%Number of params: 26MGFLOPs: 55.6
instance-segmentation-on-coco-minivalXCiT-M24/8#57mask AP: 43.7
instance-segmentation-on-coco-minivalXCiT-S24/8#60mask AP: 43.0
object-detection-on-coco-minivalXCiT-M24/8#90box AP: 48.5
object-detection-on-coco-minivalXCiT-S24/8#92box AP: 48.1
semantic-segmentation-on-ade20kXCiT-M24/8 (UperNet)#142Validation mIoU: 48.4
semantic-segmentation-on-ade20kXCiT-S24/8 (UperNet)#147Validation mIoU: 48.1
semantic-segmentation-on-ade20kXCiT-S24/8 (Semantic-FPN)#163Validation mIoU: 47.1
semantic-segmentation-on-ade20kXCiT-M24/8 (Semantic-FPN)#165Validation mIoU: 46.9
semantic-segmentation-on-ade20kXCiT-S12/8 (UperNet)#169Validation mIoU: 46.6
semantic-segmentation-on-ade20kXCiT-S12/8 (Semantic-FPN)#197Validation mIoU: 44.2