Image Classification on Places365-LT
50.9Accuracy (Many-shot)debLoRA
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| debLoRAMode=debLoRA adaptation2024.10 | 50.9 | — | 51.2 | 49.2 | |
| LoRAMode=Low-Rank Adaptation2024.10 | 48.2 | — | 42 | 44.9 | |
| BaselineBackbone=ResNet-1522021.03 | 45.7 | 30.2 | 27.3 | 8.2 | |
| Basel. (CE)Backbone=ResNet152, Pre-training=ImageNet pre-trained2020.08 | 45.7 | 30.2 | 27.3 | 8.2 | |
| OLTRBackbone=ResNet-1522021.03 | 44.7 | 35.9 | 37 | 25.3 | |
| OLTRBackbone=ResNet152, Pre-training=ImageNet pre-trained2020.08 | 44.7 | 35.9 | 37 | 25.3 | |
| Fine-TuneMode=Fine-Tuning2024.10 | 43.2 | — | 31.1 | 39 | |
| Feature AugBackbone=ResNet-1522021.03 | 42.8 | 36.4 | 37.5 | 22.7 | |
| FSABackbone=ResNet152, Pre-training=ImageNet pre-trained2020.08 | 42.8 | 36.4 | 37.5 | 22.7 | |
| CRTBackbone=ResNet-1522021.03 | 42 | 36.7 | 37.6 | 24.9 | |
| Decouple-cRTBackbone=ResNet152, Pre-training=ImageNet pre-trained2020.08 | 42 | 36.7 | 37.6 | 24.9 | |
| Focal LossBackbone=ResNet-1522021.03 | 41.1 | 34.6 | 34.8 | 22.4 | |
| Range LossBackbone=ResNet-1522021.03 | 41.1 | 35.1 | 35.4 | 23.2 | |
| Lifted LossBackbone=ResNet152, Pre-training=ImageNet pre-trained2020.08 | 41.1 | 35.2 | 35.4 | 24 | |
| Focal LossBackbone=ResNet152, Pre-training=ImageNet pre-trained2020.08 | 41.1 | 34.6 | 34.8 | 22.4 | |
| LWSBackbone=ResNet-1522021.03 | 40.6 | 37.6 | 39.1 | 28.6 | |
| Decouple-LWSBackbone=ResNet152, Pre-training=ImageNet pre-trained2020.08 | 40.6 | 37.6 | 39.1 | 28.6 | |
| NCMBackbone=ResNet-1522021.03 | 40.4 | 36.4 | 37.1 | 27.3 | |
| DisAlignBackbone=ResNet-1522021.03 | 40.4 | 39.3 | 42.4 | 30.1 | |
| Decouple-NCMBackbone=ResNet152, Pre-training=ImageNet pre-trained2020.08 | 40.4 | 36.4 | 37.1 | 27.3 | |
| Zero-ShotMode=Zero-Shot2024.10 | 40.3 | — | 36.9 | 24.9 | |
| MBJBackbone=ResNet152, Pre-training=ImageNet pre-trained2020.08 | 39.5 | 38.1 | 38.2 | 35.5 | |
| tau-normBackbone=ResNet-1522021.03 | 37.8 | 37.9 | 40.7 | 31.8 | |
| Decouple-tau-normBackbone=ResNet152, Pre-training=ImageNet pre-trained2020.08 | 37.8 | 37.9 | 40.7 | 31.8 |