Deep Learning for Computer Vision

  1. Classical CNN Architectures
    1. LeNet Family
      1. LeNet-1
        1. Historical Context
        2. LeNet-5
          1. Architecture Details
            1. Convolutional Layers
              1. Subsampling Layers
                1. Fully Connected Layers
                  1. Activation Functions
                    1. Applications and Impact
                  2. AlexNet
                    1. Architectural Innovations
                      1. Deep Architecture
                        1. ReLU Activation
                          1. Dropout Regularization
                            1. Data Augmentation
                              1. GPU Implementation
                              2. Layer-by-layer Analysis
                                1. Convolutional Layers
                                  1. Pooling Layers
                                    1. Fully Connected Layers
                                    2. Training Details
                                      1. Optimization Strategy
                                        1. Learning Rate Schedule
                                        2. ImageNet Performance
                                          1. Error Rate Reduction
                                            1. Competition Impact
                                          2. VGG Networks
                                            1. Design Philosophy
                                              1. Small Filter Size
                                                1. Deep Architecture
                                                  1. Uniform Structure
                                                  2. VGG-11
                                                    1. Architecture Overview
                                                    2. VGG-13
                                                      1. Additional Depth
                                                      2. VGG-16
                                                        1. Detailed Architecture
                                                          1. Layer Configuration
                                                          2. VGG-19
                                                            1. Maximum Depth Variant
                                                            2. Computational Analysis
                                                              1. Parameter Count
                                                                1. Memory Requirements
                                                                  1. Inference Speed
                                                                  2. Applications and Variants
                                                                    1. Feature Extraction
                                                                      1. Style Transfer
                                                                    2. GoogLeNet and Inception
                                                                      1. Inception Module Design
                                                                        1. Multi-scale Processing
                                                                          1. Parallel Convolutions
                                                                            1. Dimensionality Reduction
                                                                              1. 1x1 Convolutions
                                                                              2. GoogLeNet Architecture
                                                                                1. Inception Modules
                                                                                  1. Auxiliary Classifiers
                                                                                    1. Global Average Pooling
                                                                                    2. Inception v2
                                                                                      1. Batch Normalization
                                                                                        1. Factorized Convolutions
                                                                                        2. Inception v3
                                                                                          1. Asymmetric Convolutions
                                                                                            1. Label Smoothing
                                                                                            2. Inception v4
                                                                                              1. Residual Connections
                                                                                              2. Inception-ResNet
                                                                                                1. Hybrid Architecture
                                                                                              3. ResNet Family
                                                                                                1. Motivation
                                                                                                  1. Degradation Problem
                                                                                                    1. Deep Network Training
                                                                                                    2. Residual Learning
                                                                                                      1. Skip Connections
                                                                                                        1. Identity Mappings
                                                                                                          1. Gradient Flow
                                                                                                          2. Residual Block Design
                                                                                                            1. Basic Block
                                                                                                              1. Bottleneck Block
                                                                                                                1. Pre-activation Blocks
                                                                                                                2. ResNet Variants
                                                                                                                  1. ResNet-18
                                                                                                                    1. ResNet-34
                                                                                                                      1. ResNet-50
                                                                                                                        1. ResNet-101
                                                                                                                          1. ResNet-152
                                                                                                                          2. Wide ResNet
                                                                                                                            1. Width vs Depth
                                                                                                                              1. Computational Efficiency
                                                                                                                              2. ResNeXt
                                                                                                                                1. Cardinality Dimension
                                                                                                                                  1. Group Convolutions
                                                                                                                                2. DenseNet
                                                                                                                                  1. Dense Connectivity
                                                                                                                                    1. Feature Reuse
                                                                                                                                      1. Gradient Flow
                                                                                                                                      2. Dense Block Structure
                                                                                                                                        1. Growth Rate
                                                                                                                                          1. Bottleneck Layers
                                                                                                                                          2. Transition Layers
                                                                                                                                            1. Dimensionality Reduction
                                                                                                                                              1. Pooling Operations
                                                                                                                                              2. DenseNet Variants
                                                                                                                                                1. DenseNet-121
                                                                                                                                                  1. DenseNet-169
                                                                                                                                                    1. DenseNet-201
                                                                                                                                                    2. Memory Efficiency
                                                                                                                                                      1. Implementation Considerations

                                                                                                                                                  Previous

                                                                                                                                                  3. Training Deep Vision Models

                                                                                                                                                  Go to top

                                                                                                                                                  Next

                                                                                                                                                  5. Modern CNN Architectures