ML-Decoder: Scalable and Versatile Classification Head

Benchmark Model Rank Results
fine-grained-image-classification-on-stanfordTResNet-L + ML-Decoder#4Accuracy: 96.41%
image-classification-on-cifar-100Swin-L + ML-Decoder#2Percentage correct: 95.1
multi-label-classification-on-ms-cocoML-Decoder(TResNet-XL, resolution 640)#1mAP: 91.4
multi-label-classification-on-ms-cocoML-Decoder(TResNet-L, resolution 640)#4mAP: 91.1
multi-label-classification-on-openimages-v6TResNet-M#2mAP: 86.8
multi-label-zero-shot-learning-on-nus-wideML-Decoder#3mAP: 31.1