Deep Learning for Computer Vision

  1. Training Deep Vision Models
    1. Dataset Preparation
      1. Common Vision Datasets
        1. MNIST
          1. Handwritten Digit Recognition
            1. Dataset Characteristics
              1. Baseline Performance
              2. Fashion-MNIST
                1. Clothing Item Classification
                2. CIFAR-10
                  1. Natural Image Classification
                    1. Dataset Challenges
                    2. CIFAR-100
                      1. Fine-grained Classification
                      2. ImageNet
                        1. Large-scale Visual Recognition
                          1. Dataset Statistics
                            1. Evaluation Protocol
                            2. COCO
                              1. Object Detection
                                1. Instance Segmentation
                                  1. Keypoint Detection
                                    1. Captioning Tasks
                                    2. Pascal VOC
                                      1. Object Detection and Segmentation
                                      2. Open Images
                                        1. Large-scale Object Detection
                                        2. CelebA
                                          1. Face Attribute Recognition
                                        3. Data Loading and Preprocessing
                                          1. Image Loading
                                            1. File Format Handling
                                              1. Memory Management
                                              2. Resizing Strategies
                                                1. Aspect Ratio Preservation
                                                  1. Center Cropping
                                                    1. Padding Methods
                                                    2. Normalization Techniques
                                                      1. Pixel Value Scaling
                                                        1. Channel-wise Normalization
                                                          1. Dataset Statistics
                                                          2. Data Type Conversion
                                                            1. Float vs Integer Representations
                                                          3. Data Augmentation
                                                            1. Geometric Transformations
                                                              1. Rotation
                                                                1. Angle Selection
                                                                  1. Interpolation Methods
                                                                  2. Scaling and Zooming
                                                                    1. Scale Factor Selection
                                                                    2. Translation
                                                                      1. Boundary Handling
                                                                      2. Shearing
                                                                        1. Perspective Transformation
                                                                          1. Elastic Deformation
                                                                          2. Photometric Transformations
                                                                            1. Brightness Adjustment
                                                                              1. Contrast Modification
                                                                                1. Saturation Changes
                                                                                  1. Hue Shifting
                                                                                    1. Gamma Correction
                                                                                    2. Noise Addition
                                                                                      1. Gaussian Noise
                                                                                        1. Salt and Pepper Noise
                                                                                          1. Speckle Noise
                                                                                          2. Occlusion Techniques
                                                                                            1. Random Erasing
                                                                                              1. Cutout
                                                                                                1. Hide-and-Seek
                                                                                                2. Advanced Augmentation
                                                                                                  1. Mixup
                                                                                                    1. Linear Interpolation
                                                                                                      1. Label Smoothing
                                                                                                      2. CutMix
                                                                                                        1. Spatial Mixing
                                                                                                        2. AutoAugment
                                                                                                        3. RandAugment
                                                                                                          1. Simplified Policy
                                                                                                      3. Data Splitting Strategies
                                                                                                        1. Training Set
                                                                                                          1. Validation Set
                                                                                                            1. Test Set
                                                                                                              1. Cross-validation
                                                                                                                1. Stratified Sampling
                                                                                                              2. Loss Functions for Vision Tasks
                                                                                                                1. Classification Losses
                                                                                                                  1. Binary Cross-Entropy
                                                                                                                    1. Sigmoid Activation
                                                                                                                      1. Class Imbalance Handling
                                                                                                                      2. Categorical Cross-Entropy
                                                                                                                        1. Softmax Activation
                                                                                                                          1. Multi-class Classification
                                                                                                                          2. Sparse Categorical Cross-Entropy
                                                                                                                            1. Integer Label Handling
                                                                                                                            2. Focal Loss
                                                                                                                              1. Hard Example Mining
                                                                                                                                1. Class Imbalance Solutions
                                                                                                                                2. Label Smoothing
                                                                                                                                  1. Overconfidence Prevention
                                                                                                                                3. Regression Losses
                                                                                                                                  1. Mean Squared Error
                                                                                                                                    1. L2 Loss Properties
                                                                                                                                    2. Mean Absolute Error
                                                                                                                                      1. L1 Loss Properties
                                                                                                                                        1. Robustness to Outliers
                                                                                                                                        2. Huber Loss
                                                                                                                                          1. Smooth L1 Loss
                                                                                                                                            1. Combining L1 and L2
                                                                                                                                          2. Segmentation Losses
                                                                                                                                            1. Dice Loss
                                                                                                                                              1. Overlap Measurement
                                                                                                                                                1. Class Imbalance Handling
                                                                                                                                                2. Intersection over Union Loss
                                                                                                                                                  1. Jaccard Index
                                                                                                                                                  2. Tversky Loss
                                                                                                                                                    1. Generalized Dice Loss
                                                                                                                                                    2. Boundary Loss
                                                                                                                                                      1. Edge-aware Segmentation
                                                                                                                                                    3. Detection Losses
                                                                                                                                                      1. Smooth L1 Loss
                                                                                                                                                        1. Bounding Box Regression
                                                                                                                                                        2. Balanced Cross-Entropy
                                                                                                                                                          1. Positive-Negative Balance
                                                                                                                                                      2. Regularization Techniques
                                                                                                                                                        1. Weight Regularization
                                                                                                                                                          1. L1 Regularization
                                                                                                                                                            1. Sparsity Promotion
                                                                                                                                                              1. Feature Selection
                                                                                                                                                              2. L2 Regularization
                                                                                                                                                                1. Weight Decay
                                                                                                                                                                  1. Smooth Solutions
                                                                                                                                                                  2. Elastic Net
                                                                                                                                                                    1. L1 and L2 Combination
                                                                                                                                                                  3. Architectural Regularization
                                                                                                                                                                    1. Dropout Variants
                                                                                                                                                                      1. Standard Dropout
                                                                                                                                                                        1. Spatial Dropout
                                                                                                                                                                          1. DropConnect
                                                                                                                                                                          2. Batch Normalization Effects
                                                                                                                                                                            1. Implicit Regularization
                                                                                                                                                                          3. Training Regularization
                                                                                                                                                                            1. Early Stopping
                                                                                                                                                                              1. Validation Monitoring
                                                                                                                                                                                1. Patience Parameter
                                                                                                                                                                                2. Learning Rate Scheduling
                                                                                                                                                                                  1. Regularization Effects
                                                                                                                                                                                3. Data-based Regularization
                                                                                                                                                                                  1. Data Augmentation
                                                                                                                                                                                    1. Noise Injection
                                                                                                                                                                                      1. Adversarial Training
                                                                                                                                                                                    2. Transfer Learning and Fine-tuning
                                                                                                                                                                                      1. Pre-trained Model Selection
                                                                                                                                                                                        1. Model Zoo Resources
                                                                                                                                                                                          1. Architecture Considerations
                                                                                                                                                                                            1. Domain Similarity Assessment
                                                                                                                                                                                            2. Feature Extraction Approach
                                                                                                                                                                                              1. Frozen Feature Extractor
                                                                                                                                                                                                1. Layer Selection
                                                                                                                                                                                                  1. Feature Dimensionality
                                                                                                                                                                                                  2. Fine-tuning Strategies
                                                                                                                                                                                                    1. Full Network Fine-tuning
                                                                                                                                                                                                      1. Layer-wise Fine-tuning
                                                                                                                                                                                                        1. Gradual Unfreezing
                                                                                                                                                                                                          1. Learning Rate Adjustment
                                                                                                                                                                                                          2. Domain Adaptation
                                                                                                                                                                                                            1. Domain Shift Challenges
                                                                                                                                                                                                              1. Unsupervised Domain Adaptation
                                                                                                                                                                                                                1. Few-shot Domain Adaptation
                                                                                                                                                                                                                2. Multi-task Learning
                                                                                                                                                                                                                  1. Shared Representations
                                                                                                                                                                                                                    1. Task-specific Heads
                                                                                                                                                                                                                      1. Loss Balancing

                                                                                                                                                                                                                  Previous

                                                                                                                                                                                                                  2. Convolutional Neural Networks

                                                                                                                                                                                                                  Go to top

                                                                                                                                                                                                                  Next

                                                                                                                                                                                                                  4. Classical CNN Architectures