Please enable JavaScript.
Coggle requires JavaScript to display documents.
Deep Learning with MATLAB, ImageNet - Coggle Diagram
Deep Learning with MATLAB
Chp2 Transfer Learning
For Image Classification
Transfer Learning
1000 cat to 14 cat
Prepare Dataset
Augment
mirror
translation
resize
make sure it is the same inputsize of the network
flipping
Folder, categories are in separate folders
remove the last 3 layers
last FC
number of categories
last Classification
trainingoptions
minibatchsize
initial learningrate
epochs
solvers
training enviroment
CPU
GPU
DeepNetwork Desiger
Pretrained
Load
add - on
ONNX.ai
Tensorflow
Pytorch
Keras
.mat
Types
CNNs
Googlenet
dag network
Resnet
Alexnet
series network
LSTMs
Things to look out
Input Image Size
227x227x3
224x224x3
300x300x3
Chp 3 Interpreting
Network Behavior
Activations
ReLu
Max Pooling
Convolution Filters
First Layers
Simple Features
Color
Shape
Edge
Deep Layers
Fur
Facial Features
Eyes
Occlusion Sensitivity
ImageLime
Feature Extraction
First Layers Up Until the last FC
Is Auto Feature Extraction
Machine Learning Model
Trees
SVMs
Fully Connect Layer (NN)
Regressions
Logistics
Polynomial
Chap 4 Creating
the Network
Pretrained
1000 Categories
Image Net
TransferLearning
Hyperparameters
Solver
LearningRate
Minibatchsize
Number of Categories
Training Dataset (Scope)
Train from Scratch
Architecture
Conv
Size
Stride
Filters
Padding
Relu
Max Pooling
BatchNormalization
Series
Alexnet
Dag
Resnet50
Things to take Note
Number of Categories
Image Input Size
Follow Reference Models -> Layer patterns
Chap5 Training the Network
Monitor
Validation Plot
Nothing Change
Hyperparemeters
Datasets
Architecture
Explosion of accuracy
Dataset
Split
Test, it has not seen
during training
Training
Training
Validation
Network training
Initial Learning Rate
fixed
scheduled
exploration
exploitation
Mini Batch Size
Approximation Loss Function
Minimal
Maximum
Solver
sgdm
adam
reg
Chp 6 Improving Network
Performance
Experiments
Experiment Manager
Design of Experiments
Grid Search
Random Search
Bayesian Optimization
Training Options
Epochs
How many times we pass through the dataset
If not flatted out then increase value
It means we can still get information to modify Learnables
Minibatch Size
Initial Learning Rate
L2 Regularization
Weights Decay
Learning Rate Schedule
Exploration / Exploitation
Augmenting Images
Translation / Resize / Mirror / Flip / Translation X and Y
Synthesize Data using AI
Digital Twins
Balanced Data
Chp 7 Performing
Image Regression
Classification
Discrete
Categories
Buckets
Regression
Conti
Forecasting
Timeseries
Scalar
Transfer Learning
Replace the Layers
Fully Connected Layer
RGB == 3 Channels
Angle == 1 Channel
Regression Layer
Machine learning
Preprocessing helps
Normalization
Removing Outliers
Filling Missing Data
Translation / Interpretation
Training from Scratch
Pretrained
Chp 8 Deep Learning
with Computer Vision
Localization
RCNN
Others
Yolo
v2
Base Network
Alexnet
Resnet50
Restnet18
YoloV2Layers
Anchorboxes
Train / Detection
Scores
Labels
Bounding box
v4
v3
IPCV
Extract
Match
Detect
Bounding box
Clustering
Semantic Segmentation
Classification
Image to Image Regression
Video Classification
Labeling Apps
Video Labeler
Automation
Point Tracking / IPCV
FAST
BRISK
MSER
ORB
ETC
Built in
ACF Car
ACF People
Surgeon 15 years -> Clinical Research
Manual
Signal / Audio Labeler
Image Labeler
Lidar Labeler
Ground Truth Labeler
for Autonomous Driving
LSTMs
Classifiation
Channels
Text
Vocab, all letters, all words
51 (Unique)
Letters
Spaces
Caps
2000 Buffer
Signals
Time Series
Sequence / Regression
Matrix -> Vector
Matrix -> Matrix
Matrix -> Scalar
Vector -> Vector
Vector -> Matrix
Vector -> Scalar
Images
Input Layer
TimeSeries
Scalar
ImageNet