Multi-label classification on COCO (test val)
91.1mAPML-Decoder
Evaluation results
| Method | Links | |
|---|---|---|
| ML-DecoderBackbone=TResNetL, Resolution=640 x 640, Pre-trained=ImageNet-21K2022.09 | 91.1 | |
| Query2LabelBackbone=TResNetL, Resolution=640 x 640, Pre-trained=ImageNet-21K2022.09 | 90.3 | |
| MiTr-XLResolution=384 x 384, Pre-trained=OpenImage2022.09 | 90 | |
| Obj2SeqBackbone=ResNet50, Resolution=800 x 13332022.09 | 89 | |
| Obj2SeqBackbone=ResNet50, Resolution=480 x 6402022.09 | 87 | |
| ASLBackbone=TResNetL, Resolution=448 x 4482022.09 | 86.6 | |
| Query2LabelBackbone=ResNet101, Resolution=448 x 4482022.09 | 84.9 | |
| MCARBackbone=ResNet101, Resolution=576 x 5762022.09 | 84.5 | |
| MCARBackbone=ResNet101, Resolution=448 x 4482022.09 | 83.8 | |
| ResNet-101 + ReLabel (oracle)Backbone=ResNet-101, Input size=448x448, Training strategy=ReLabel (oracle/segmentation ground truth)2021.01 | 80.9 | |
| ResNet-101 + ReLabel (machine)Backbone=ResNet-101, Input size=448x448, Training strategy=ReLabel (machine-generated label maps)2021.01 | 79 | |
| ResNet-101Backbone=ResNet-101, Input size=448x4482021.01 | 76.6 | |
| ResNet-50 + ReLabel (oracle)Backbone=ResNet-50, Input size=224x224, Training strategy=ReLabel (oracle/segmentation ground truth)2021.01 | 73.2 | |
| ResNet-50 + ReLabel (machine)Backbone=ResNet-50, Input size=224x224, Training strategy=ReLabel (machine-generated label maps)2021.01 | 72.7 | |
| ResNet-50Backbone=ResNet-50, Input size=224x2242021.01 | 69 |