Object Detection on PASCAL VOC 2007 (test) (mAP and Compression Metrics)
74.5mAPFull-precision
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Full-precisionArchitecture=Faster R-CNN, Backbone=ResNet-18, Bit-width (W/A)=32/322023.03 | 74.5 | 1 | 1 | |
| Full-precisionArchitecture=SSD300, Backbone=VGG-16, Bit-width (W/A)=32/322023.03 | 72.4 | 1 | 1 | |
| SparksArchitecture=SSD300, Backbone=VGG-16, Bit-width (W/A)=0.78/1, Two-step training=true, Generalized Sign/PReLU=true2023.03 | 68.9 | 41 | 108 | |
| SparksArchitecture=SSD300, Backbone=VGG-16, Bit-width (W/A)=0.56/1, Two-step training=true, Generalized Sign/PReLU=true2023.03 | 68 | 57.1 | 285 | |
| SparksArchitecture=Faster R-CNN, Backbone=ResNet-18, Bit-width (W/A)=0.78/1, Two-step training=true, Generalized Sign/PReLU=true2023.03 | 66.2 | 41 | 88 | |
| BiDetArchitecture=SSD300, Backbone=VGG-16, Bit-width (W/A)=1/12023.03 | 66 | 32 | 64 | |
| SparksArchitecture=Faster R-CNN, Backbone=ResNet-18, Bit-width (W/A)=0.56/1, Two-step training=true, Generalized Sign/PReLU=true2023.03 | 65.5 | 57.1 | 214 | |
| SparksArchitecture=SSD300, Backbone=VGG-16, Bit-width (W/A)=0.78/1, Two-step training=false, Generalized Sign/PReLU=false2023.03 | 65.2 | 41 | 108 | |
| SparksArchitecture=SSD300, Backbone=VGG-16, Bit-width (W/A)=0.56/1, Two-step training=false, Generalized Sign/PReLU=false2023.03 | 64.3 | 57.1 | 285 | |
| Bi-RealNetArchitecture=SSD300, Backbone=VGG-16, Bit-width (W/A)=1/12023.03 | 63.8 | 32 | 64 | |
| BiDetArchitecture=Faster R-CNN, Backbone=ResNet-18, Bit-width (W/A)=1/12023.03 | 59.5 | 32 | 64 | |
| SparksArchitecture=Faster R-CNN, Backbone=ResNet-18, Bit-width (W/A)=0.78/1, Two-step training=false, Generalized Sign/PReLU=false2023.03 | 58.9 | 41 | 88 | |
| Bi-RealNetArchitecture=Faster R-CNN, Backbone=ResNet-18, Bit-width (W/A)=1/12023.03 | 58.2 | 32 | 64 | |
| SparksArchitecture=Faster R-CNN, Backbone=ResNet-18, Bit-width (W/A)=0.56/1, Two-step training=false, Generalized Sign/PReLU=false2023.03 | 57.7 | 57.1 | 214 | |
| XNOR-NetArchitecture=SSD300, Backbone=VGG-16, Bit-width (W/A)=1/12023.03 | 50.2 | 32 | 64 | |
| XNOR-NetArchitecture=Faster R-CNN, Backbone=ResNet-18, Bit-width (W/A)=1/12023.03 | 48.4 | 32 | 64 | |
| BNNArchitecture=SSD300, Backbone=VGG-16, Bit-width (W/A)=1/12023.03 | 42 | 32 | 64 | |
| BNNArchitecture=Faster R-CNN, Backbone=ResNet-18, Bit-width (W/A)=1/12023.03 | 35.6 | 32 | 64 |