【学习笔记】深度学习实战 | 流行深度学习网络架构

简要声明
- 学习相关网址
- [双语字幕]吴恩达深度学习deeplearning.ai
- Papers With Code
- Datasets
- 深度学习网络基于PyTorch学习架构。
- 本学习笔记单纯是为了能对学到的内容有更深入的理解,如果有错误的地方,恳请包容和指正。
参考文献
- LeNet (1998) [Gradient-based learning applied to document recognition]
- AlexNet (2012) [Imagenet classification with deep convolutional neural networks]
- VGG (2014) [Very Deep Convolutional Networks for Large-Scale Image Recognition]
- GoogleNet (2014) [Going Deeper with Convolutions]
- ResNet (2015) [Deep Residual Learning for Image Recognition]
简要介绍
LeNet-5 (1998)

| Dataset | MNIST |
|---|---|
| Input (feature maps) | 32×32 (28×28) |
| CONV Layers | 2 |
| FC Layers | 2 |
| Activation | Sigmoid |
| Output | 10 |
AlexNet (2012)

| Dataset | ImageNet |
|---|---|
| Input (feature maps) | 256×256×3 (224×224×3) |
| CONV Layers | 5 |
| FC Layers | 3 |
| Activation | ReLU |
| Output | 1000 |
VGG-16 (2014)

| Dataset | ImageNet |
|---|---|
| Input (feature maps) | 256×256×3 (224×224×3) |
| CONV Layers | 13 |
| FC Layers | 3 |
| Activation | ReLU |
| Output | 1000 |
GoogLeNet (2014)

| Dataset | ImageNet |
|---|---|
| Input (feature maps) | 256×256×3 (224×224×3) |
| CONV Layers | 21(depth), 57(total) |
| FC Layers | 1 |
| Activation | ReLU |
| Output | 1000 |
ResNet-50 (2015)

| Dataset | ImageNet |
|---|---|
| Input (feature maps) | 256×256×3 (224×224×3) |
| CONV Layers | 49 |
| FC Layers | 1 |
| Activation | ReLU |
| Output | 1000 |
Summary
| LeNet-5 | AlexNet | VGG-16 | GoogLeNet | ResNet-50 | |
|---|---|---|---|---|---|
| Dataset | MNIST | ImageNet | ImageNet | ImageNet | ImageNet |
| Input (feature maps) | 32×32 (28×28) | 256×256×3 (224×224×3) | 256×256×3 (224×224×3) | 256×256×3 (224×224×3) | 256×256×3 (224×224×3) |
| CONV Layers | 2 | 5 | 16 | 21(depth) | 49 |
| FC Layers | 2 | 3 | 3 | 1 | 1 |
| *Total Weight | 60k | 61M | 138M | 7M | 25.5M |
| *Total MACs | 341k | 724M | 15.5G | 1.43G | 3.9G |