Multi-label Classification on PASCAL VOC 2007/2012 (test)
95.48Macro AUCCentralized
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CentralizedBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Optimizer=AdamW, Batch Size=322025.09 | 95.48 | 74.61 | 96.11 | 76.94 | |
| CentralizedBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Optimizer=AdamW2025.09 | 95.48 | 74.61 | 96.11 | 76.94 | |
| FedNCA-MLBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Communication Rounds=100, Local Epochs per round=1, Batch Size=32, Optimizer=AdamW, Initial Learning Rate=1 x 10^-4, Weight Decay=0.01, Negative feature rejection coefficient (lambda_1)=0.01, Positive feature contrastive coefficient (lambda_2)=12025.09 | 93.82 | 64.28 | 94.52 | 67.61 | |
| FedProxBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Communication Rounds=100, Local Epochs per round=1, Batch Size=32, Optimizer=AdamW, Negative feature rejection coefficient (lambda_1)=0.01, Positive feature contrastive coefficient (lambda_2)=12025.09 | 93.55 | 57.04 | 94.04 | 64.43 | |
| FedAvgBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Communication Rounds=100, Local Epochs per round=1, Batch Size=32, Optimizer=AdamW, Initial Learning Rate=1 x 10^-4, Weight Decay=0.01, Negative feature rejection coefficient (lambda_1)=0.01, Positive feature contrastive coefficient (lambda_2)=12025.09 | 93.54 | 57.57 | 94.13 | 64.67 | |
| FedCurvBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Communication Rounds=100, Local Epochs per round=1, Batch Size=32, Optimizer=AdamW, Negative feature rejection coefficient (lambda_1)=0.01, Positive feature contrastive coefficient (lambda_2)=12025.09 | 93.53 | 57.36 | 94.1 | 64.6 | |
| SCAFFOLDBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Communication Rounds=100, Local Epochs per round=1, Batch Size=32, Optimizer=AdamW, Negative feature rejection coefficient (lambda_1)=0.01, Positive feature contrastive coefficient (lambda_2)=12025.09 | 93.41 | 49.64 | 93.6 | 61.44 | |
| FedAvgBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Communication Rounds=100, Local Epochs per round=1, Batch Size=32, Optimizer=AdamW, Initial Learning Rate=1 x 10^-4, Weight Decay=0.01, Negative feature rejection coefficient (lambda_1)=0.01, Positive feature contrastive coefficient (lambda_2)=12025.09 | 93.31 | 47.73 | 93.05 | 60.46 | |
| SCAFFOLDBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Communication Rounds=100, Local Epochs per round=1, Batch Size=32, Optimizer=AdamW, Negative feature rejection coefficient (lambda_1)=0.01, Positive feature contrastive coefficient (lambda_2)=12025.09 | 93.17 | 57.44 | 94.03 | 64.95 | |
| FedProxBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Communication Rounds=100, Local Epochs per round=1, Batch Size=32, Optimizer=AdamW, Negative feature rejection coefficient (lambda_1)=0.01, Positive feature contrastive coefficient (lambda_2)=12025.09 | 93.14 | 49.49 | 93.8 | 62.18 | |
| FedCurvBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Communication Rounds=100, Local Epochs per round=1, Batch Size=32, Optimizer=AdamW, Negative feature rejection coefficient (lambda_1)=0.01, Positive feature contrastive coefficient (lambda_2)=12025.09 | 93.13 | 48.54 | 93.81 | 60.49 | |
| FedNCA-MLBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Communication Rounds=100, Local Epochs per round=1, Batch Size=32, Optimizer=AdamW, Initial Learning Rate=1 x 10^-4, Weight Decay=0.01, Negative feature rejection coefficient (lambda_1)=0.01, Positive feature contrastive coefficient (lambda_2)=12025.09 | 93.01 | 61.08 | 93.7 | 65.05 | |
| FedLGTBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Communication Rounds=100, Local Epochs per round=1, Batch Size=32, Optimizer=AdamW, Negative feature rejection coefficient (lambda_1)=0.01, Positive feature contrastive coefficient (lambda_2)=12025.09 | 91.93 | 62.11 | 91.75 | 67.58 | |
| FedLGTBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Communication Rounds=100, Local Epochs per round=1, Batch Size=32, Optimizer=AdamW, Negative feature rejection coefficient (lambda_1)=0.01, Positive feature contrastive coefficient (lambda_2)=12025.09 | 91.25 | 56.53 | 91.46 | 63.51 | |
| SphereFedBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Communication Rounds=100, Local Epochs per round=1, Batch Size=32, Optimizer=AdamW, Negative feature rejection coefficient (lambda_1)=0.01, Positive feature contrastive coefficient (lambda_2)=12025.09 | 84.25 | 32.44 | 85.94 | 35.18 | |
| SphereFedBackbone=ResNet-18, Pre-training=ImageNet-pretrained weights, Communication Rounds=100, Local Epochs per round=1, Batch Size=32, Optimizer=AdamW, Negative feature rejection coefficient (lambda_1)=0.01, Positive feature contrastive coefficient (lambda_2)=12025.09 | 83.72 | 33.49 | 84.38 | 38.19 |