CNN
- Convolutional Neural Network (aka CNN, ConvNet)
- Deep neural networks commonly applied to analyzing visual imagery
- Applications: Image recognition | Video recognition | Image classification | NLP
- State-of-the-art : AlexNet | VGG | ResNet | Inception | EfficientNet
- Others: DenseNet | MobileNet | ResNext
| Name | Year | Top-1 accuracy | Top-5 accuracy | Layers | Parameters | FLOP(s) |
|---|---|---|---|---|---|---|
| AlexNet | 2012 | 57.0% | 80.3% | 8 | 60M | - |
| VGG-16 | 2014 | 71.3% | 90.1% | 16 | 138M | - |
| VGG-19 | 2014 | 71.3% | 90.0% | 19 | 143M | - |
| Inception-v1 | 2014 | 5M | - | |||
| Inception-v3 | 2016 | 78.8% | 94.4% | 24M | 5.7B | |
| ResNet-50 | 2016 | 76.0% | 93.0% | 50 | 26M | 4.1B |
| ResNet-152 | 2016 | 77.8% | 93.8% | 60M | 11B | |
| DenseNet-169 | 2017 | 76.2% | 93.2% | 14M | 3.5B | |
| DenseNet-264 | 2017 | 77.9% | 93.9% | 34M | 6.0B | |
| Xception | 2017 | 79.0% | 94.5% | 23M | 8.4B | |
| ResNeXt-101 | 2017 | 80.9% | 95.6% | 84M | 32B | |
| Inception-v4 | 2017 | 80.0% | 95.0% | 48M | 13B | |
| MobileNet | 2017 | 70.4% | 89.5% | 4.2M | - | |
| Inception-resnet-v2 | 2017 | 80.1% | 95.1% | 56M | 13B | |
| MobileNetV2 | 2018 | 71.3% | 90.1% | 3.5M | - | |
| EfficientNet-B0 | 2019 | 77.3% | 93.5% | 5.3M | 0.39B | |
| EfficientNet-B1 | 2019 | 79.2% | 94.5% | 7.8M | 0.70B | |
| EfficientNet-B2 | 2019 | 80.3% | 95.0% | 9.2M | 1.0B | |
| EfficientNet-B3 | 2019 | 81.7% | 95.6% | 12M | 1.8B | |
| EfficientNet-B4 | 2019 | 83.0% | 96.3% | 19M | 4.2B | |
| EfficientNet-B5 | 2019 | 83.7% | 96.7% | 30M | 9.9B | |
| EfficientNet-B6 | 2019 | 84.2% | 96.8% | 43M | 19B | |
| EfficientNet-B7 | 2019 | 84.4% | 97.1% | 66M | 37B |
