多尺度上下文引导特征消除的古塔图像分类
|
孟月波,王博,刘光辉
|
Multi-scale context-guided feature elimination for ancient tower image classification
|
Yuebo MENG,Bo WANG,Guanghui LIU
|
|
表 7 不同算法在细粒度数据集上的准确率对比 |
Tab.7 Comparison of accuracy of different algorithms on fine-grained datasets |
|
方法 | 主干网络 | 分辨率 | P/% | CUB-200-2011 | Stanford Cars | Aircraft | WS-DAN[5] | Inception v3 | 448×448 | 89.4 | 94.5 | 93.0 | PMG[7] | ResNet-50 | 550×550 | 89.6 | 95.1 | 93.4 | API-Net[3] | DenseNet-161 | 512×512 | 90.0 | 95.3 | 93.9 | PART[21] | ResNet-101 | 448×448 | 90.1 | 95.3 | 94.6 | CAL[9] | ResNet-101 | 448×448 | 90.6 | 95.5 | 94.2 | FFVT[12] | ViT-B_16 | 448×448 | 91.6 | 94.1 | 94.3 | TransFG[13] | ViT-B_16 | 448×448 | 91.7 | 94.8 | 94.1 | CAP[4] | Xception | 224×224 | 91.8 | 95.7 | 94.5 | ViT-SAC[16] | ViT-B_16 | 448×448 | 91.8 | 95.0 | 93.1 | DCAL[20] | R50-ViT-Base | 448×448 | 92.0 | 95.3 | 93.3 | 本研究方法 | MogaNet-L | 224×224 | 92.4 | 95.3 | 94.6 |
|
|
|