图像分类数据集的不同网络的成绩参考及文章出处
目录
问题
数据集
编者注：
参考文献
问题
我们在对比不同的方法和数据集时，不太好找到原参考文献。我在搜寻的时候，需要去一篇一篇的文章找，也不是特别方便，故整理于此，方便后续查看，该文章会阶段性地更新内容。
数据集
目前包括的数据集有ImageNet[1]，CIFAR（100和10）[2]。
| paper | methods | CIFAR-10 |
CIFAR-100 |
ImageNet (224x224） |
|
|---|---|---|---|---|---|
| Top-1 | Top-1 | Top-1 | Top-5 | ||
| BlueprintConv[3] | ResNet-20 | 92.2 | 67.7 | ||
| ResNet-110(BSConv-U) | 92.9 | 70.8 | |||
| MobileNetV1(x0.25) | 90.4 |
67.5 | 51.8 | ||
| MobileNetV1(x0.5) | 91.8 | 70.8 | 63.5 | 84.9 | |
| MobileNetV1(x0.75) | 92.7 | 72.2 | 68.2 | ||
| MobileNetV1(x1.0) | 93.4 | 73.4 | 70.8 | ||
| MobileNetV2(x0.25) | 89.6 | 65.6 | |||
| MobileNetV2(x0.5) | 92.0 | 72.5 | |||
| MobileNetV2(x0.75) | 93.1 | 73.2 | |||
| MobileNetV2(x1.0) | 93.6 | 74.9 | 69.7 | ||
| MobileNetV3-small(x0.35) | 90.3 | 66.5 | |||
| MobileNetV3-small(x0.5) | 91.5 | 69.4 | |||
| MobileNetV3-small(x0.75) | 92.0 | 70.4 | 65.4 | - | |
| MobileNetV3-small(x1.0) | 92.2 | 72.2 | 64.4 | ||
| MobileNetV3-large(x0.35) | 92.8 | 71.5 | |||
| MobileNetV3-large(x0.5) | 93.0 | 72.9 | |||
| MobileNetV3-large(x0.75) | 93.7 | 73.9 | 73.3 | - | |
| MobileNetV3-large(x1.0) | 93.7 | 75.2 | 71.5 | ||
| WideResNet-40-3 | 94.9 | 75.5 | |||
| WideResNet-40-8(BSConv-U) | 95.2 | 77.6 | |||
| MobileNetV1(x0.25)-BSConv | 91.6 | 69.8 | 53.2 | - | |
| MobileNetV1(x0.5)-BSConv | 93.3 | 73.5 | 64.6 | - | |
| MobileNetV1(x0.75)-BSConv | 94.3 | 74.5 | 69.2 | - | |
| MobileNetV1(x1.0)-BSConv | 94.3 | 75.7 | 71.5 | - | |
| MobileNetV2(x0.25)-BSConv | 90.1 | 68.9 | |||
| MobileNetV2(x0.5)-BSConv | 93.2 | 73.2 | |||
| MobileNetV2(x0.75)-BSConv | 93.9 | 75.0 | |||
| MobileNetV2(x1.0)-BSConv | 94.2 | 75.8 | 69.8 | - | |
| MobileNetV3-small(x0.35)-BSConv | 90.6 | 67.2 | |||
| MobileNetV3-small(x0.5)-BSConv | 91.7 | 69.6 | |||
| MobileNetV3-small(x0.75)-BSConv | 92.5 | 72.0 | |||
| MobileNetV3-small(x1.0)-BSConv | 92.7 | 73.7 | 64.8 | - | |
| MobileNetV3-large(x0.35)-BSConv | 93.0 | 73.7 | |||
| MobileNetV3-large(x0.5)-BSConv | 93.9 | 75.3 | |||
| MobileNetV3-large(x0.75)-BSConv | 94.4 | 77.0 | |||
| MobileNetV3-large(x1.0)-BSConv | 94.6 | 77.7 | 71.5 | - | |
| Ghost[4] | VGG-16 | 93.6 | |||
| l1-VGG-16 | 93.4 | ||||
| SBP-VGG-16 | 92.5 | ||||
| Ghost-VGG-16 | 93.7 | ||||
| ResNet-56 | 93.0 | ||||
| CP-ResNet-56 | 92.0 | ||||
| L1-ResNet-56 | 92.5 | ||||
| AMC-ResNet-56 | 91.9 | ||||
| Ghost-ResNet-56(s=2) | 92.7 | ||||
| ResNet-50 | 75.3 | 92.2 | |||
| Thinet-ResNet-50 | 72.1 | 90.3 | |||
| NISP-ResNet-50-B | - | 90.8 | |||
| Versatile-ResNet-50 | 74.5 | 91.8 | |||
| SSS-ResNet-50 | 74.2 | 91.9 | |||
| Ghost-ResNet-50(s=2) | 75.0 | 92.3 | |||
| Shift-ResNet-50 | 70.6 | 90.1 | |||
| Talor-FO-BN-ResNet-50 | 71.7 | - | |||
| Slimmable-ResNet-50 | 72.1 | - | |||
| MetaPruning-ResNet-50 | 73.4 | - | |||
| Ghost-ResNet-50(s=4) | 74.1 | 91.9 | |||
| ShuffleNetV1 0.5x(g=8) | 58.8 | 81.0 | |||
| MobileNetV2 0.35x | 60.3 | 82.9 | |||
| ShuffleNetV2 0.5x | 61.1 | 82.6 | |||
| GhostNet 0.5x | 66.2 | 86.6 | |||
| MobileNetV2 0.6x | 66.7 | - | |||
| ShuffleNetV1 1.0x(g=3) | 67.8 | 87.7 | |||
| ShuffleNetV2 1.0x | 69.4 | 88.9 | |||
| GhostNet 1.0x | 73.9 | 91.4 | |||
| MonileNetV2 1.0x | 71.8 | 91.0 | |||
| ShuffleNetV2 1.5x | 72.6 | 90.6 | |||
| FE-Net 1.0x | 72.9 | - | |||
| FBNet-B | 74.1 | - | |||
| ProxylessNAS | 74.6 | 92.2 | |||
| MnasNet-A1 | 75.2 | 92.5 | |||
| MobileNetV3 Large 1.0x | 75.2 |
- | |||
| GhostNet 1.3x | 75.7 | 92.7 | |||
| CBAM[5] | ResNet-18 | 70.40 | 89.45 | ||
| ResNet-18+SE | 70.59 | 89.73 | |||
| ResNet-18+CBAM | 70.73 | 89.91 | |||
| ResNet-34 | 73.31 | 91.40 | |||
| ResNet-34+SE | 73.87 | 91.65 | |||
| ResNet-34+CBAM | 74.01 | 91.76 | |||
| ResNet-50 | 75.44 | 92.50 | |||
| ResNet-50+SE | 76.86 | 93.30 | |||
| ResNet-50+CBAM | 77.34 | 93.69 | |||
| ResNet-101 | 76.62 | 93.12 | |||
| ResNet-101+SE | 77.65 | 93.81 | |||
| ResNet-101+CBAM | 78.49 | 94.31 | |||
| WideRresNet18(widen=1.5) | 73.15 | 91.12 | |||
| WideRresNet18(widen=1.5)+SE | 73.79 | 91.53 | |||
| WideRresNet18(widen=1.5)+CBAM | 73.90 | 91.57 | |||
| WideRresNet18(widen=2.0) | 74.37 | 91.80 | |||
| WideRresNet18(widen=2.0)+SE | 75.07 | 92.35 | |||
| WideRresNet18(widen=2.0)+CBAM | 75.16 | 92.37 | |||
| ResNeXt-50(32x4d) | 77.15 | 93.52 | |||
| ResNeXt-50(32x4d)+SE | 78.09 | 93.96 | |||
| ResNeXt-50(32x4d)+CBAM | 78.08 | 94.09 | |||
| ResNeXt-101(32x4d) | 78.46 | 94.25 | |||
| ResNeXt-101(32x4d)+SE | 78.83 | 94.34 | |||
| ResNeXt-101(32x4d)+CBAM | 78.93 | 94.41 | |||
| MobileNet a=0.7 | 65.14 | 86.31 | |||
| MobileNet a=0.7+SE | 67.50 | 87.51 | |||
| MobileNet a=0.7+CBAM | 68.49 | 88.52 | |||
| MobileNet | 68.61 | 88.49 | |||
| MobileNet+SE | 70.03 | 89.37 | |||
| MobileNet+CBAM | 70.99 | 90.01 | |||
| ECA[6] | |||||
编者注：
1. 若出现重名但是结果不一样的情况，见关联的文章，应该是作者自己的重复试验结果实验平台不同导致的。如MobileNetV2 1.0x。
2. CSDN的markdown编辑器不能设置合并单元格（或者跨行文本），但是这个富文本编辑器呢，也不是特别好使。markdown兼容html语言，但是我感觉那个指令挺多的，编辑起来更复杂。大家有没有简单易操作的方法推荐呢。
参考文献
[1] Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, et al. Imagenet large scale visual recognition challenge. International journal of computer vision, 115(3):211–252, 2015.
[2] Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. Technical report, Citeseer, 2009.
[3] Daniel Haase, Manuel Amthor. Rethinking Depthwise Separable Convolutions: How Intra-Kernel Correlations Lead to Improved MobileNets. In Proceedings of the IEEE conference on computer vision and pattern recognition, 2020.
[4] Kai Han, Yunhe Wang, Qi Tian, et al. GhostNet: More Features from Cheap Operation.In Proceedings of the IEEE conference on computer vision and pattern recognition, 2020.
[5] Sanghyun Woo, Jongchan Park, et al. CBAM: Convolutional Block Attention Module. In Proceedings of Europeon Conference on Computer Vision, 2018.