Please enable JavaScript.
Coggle requires JavaScript to display documents.
DEEP LEARNING FOR COMPUTER VISION - Coggle Diagram
DEEP LEARNING FOR COMPUTER VISION
COVERS
CONVOLUTIONAL NEURAL NETWORKS(CONVNETS)
PRETRAINED CONVNET
FEATURE EXTRCTION
FINE TUNNING
VISUALISING
LEARNING
CONVNETS
DATA AUGMENTATION
MITIGATE OVERFITTING
INTRODUCTION COVNETS
CONVNET
DEEP LEARNING MODEL
COMPUTER VISION
IMAGE CLASSIFICATION PROBLEMS
SMALL TRAINING DATASET
CONVOLUTION OPERATION
DENSE LAYERS
GLOBAL PATTERN
INPUT FEATURE SPACE
CONVOLUTION LAYERS
LOCAL PATTERN
PROPERTIES
LEARN
CERTAIN PATTERN
CERTAIN IMAGE POSITION
RECOGNISE THE PATTERN
ANYWHERE ELSE
SPATIAL HIERARCHIES
PATTERNS
TRANSLATION INVARIANT
FIRST CONVOLUTION LAYER
LEARN
SMALL LOCAL PATTERNS
EDGES
SECOND CONVOLUTION LAYER
LEARN
LARGER PATTERN
FEATURES
FIRST LAYERS
3D TENSORS
PATCHES
INPUT
FEATURE MAP
TRANSFORM
PATCHES
OUTPUT
FEATURE MAP
FILTERS
COLORS
(26,26,32)
BORDERING EFFECT
FEATURE MAP
5X5
OUTPUT FEATURE MAP
3X3
9 TILES
CENTER
INPUT
WIDTH
DIFFER
OUTPUT
WIDTH
HEIGHT
HEIGHT
SOLUTION
PADDING
PADDING
ADD
ROWS
COLUMNS
STRIDES
DISTANCE
TWO SUCCESSIVE WINDOWS
1
2
FEATURE MAP
WIDTH
DOWNSAMPLED
FACTOR
1 more item...
HEIGHT
EXAMPLE
LINES OF CODE
CONV2D
LAYER
MAXPOOLING2D
FIRST LAYER
INPUT TENSOR
SHAPE
IMAGE CHANNELS
IMAGE_HEIGHT
IMAGE_WIDTH
SHRINK
DEEPER
NETWORK
LAST LAYER
DENSE LAYER
CONVERT
3D VALUE
1D VECTOR
REASSEMBLED
3D OUTPUT MAP
PRETRAINED CONVNET
SUMMARY
DATA AUGMENTATION
MITIGATE
OVERFITTING
REUSE
COVNETS
NEW DATASET
FEATURE EXTRACTION
FINE-TUNING
SMALL DATASETS
FEATURE EXTRACTION
REPRESENTATIONS
PREVIOUS NETWORK
EXTRACT
FEATURES
NEW SAMPLE
RUN
NEW CLASSIFIER
MORE GENERIC
DATA AUGMENTATION
FREEZE
CONVOLUTIONAL BASE
STOP WEIGHTS
UPDATED
TWO DENSE LAYERS
WEIGHTS
UPDATED
DATASET
TOO DIFFERENT
FIRST FEW CNN LAYERS
DATA AUGMENTATION
FREEZE
CONVOLUTIONAL BASE
WEIGHT UPDATE
BEFORE COMPILATION
MODIFIED
DENSE LAYERS
RANDOMLY INITIALISED
LARGE WEIGHT UPDATES
DESTROY PRESENTATION
GENERIC MODEL
PREVIOUS DATASET
GENERAL ENOUGH
SPATIAL HIERARCHY
BIG ENOUGH
FINE-TUNING
UNFREEZE
SOME FROZEN CONV BLOCKS
SUMMARY
COVNETS
VISUAL CLASSIFICATION PROBLEM
LEARN
HIERARCHY
MODULAR PATTERNS
REPRESENT
VISUAL WORLD
CONCEPTS
REPRESENTATION
EASY
INSPECT
PRE-TRAINED
FEATURE EXTRACTION
FINE-TUNING
VISUAL DATA AUGMENTATION
FIGHT
OVERFITTING
VISUALISATION
FILTERS LEARNED
HEATMAPS
CLASS ACTIVITY
QUESTIONS
QUESTION 2
WHY
COVNETS
MORE SUCCESSFUL
DENSELY PACKED NEURAL NETWORK
QUESTION 3
DEFINITION
GLOBAL
PATTERNS
LOCAL
QUESTION 1
HOW
COVNETS
CLASSIFY
IMAGES
SEGMENT
IMAGES
LEARN
QUESTION 4
HOW
ACTIVATE
COVNET
LAYER
QUESTION 5
WHAT
ARE
MAXPOOLING
LAYERS
COV2D
QUESTION 6
HOW
PADDING
DONE
QUESTION 7
HOW
FIGHT
OVERFITTING
QUESTION 8
DEFINITION
FINE TUNING
WHY
COVNET TRAINING
OVERVIEW
TRAIN
COVNET
LITTLE DATA
OVERFITTING
PREVENT
DATA AUGMENTATION
FEATURE EXTRACTION
PRETRAINED NETWORK
FINE-TUNING
COVNETS
SMALL IMAGE DATASET
DIFFERENT PROBLEM
IMAGE CLASSIFICATION
SPEECH-TO-TEXT MODEL
TRAINED
LARGE SCALE DATASET
BUILDING NETWORK
BIGGER IAMGE
LARGER NETWORK
CONV2D + MAXPOOLING2D
AUGMENT
CAPACITY
NETWORK
REDUCE
SIZE
FEATURE MAPS
DATA PREPROCESSING
DECODE
JPEG
RGB GRID PIXELS
CONVERT
FLOATING-POINT TENSORS
READ
PICTURE FILES
RESCALE
PIXEL VALUE
0-255
[0,1] INTERVAL
DATA AUGMENTATION
OVERFITTING
TOO FEW SAMPLES
TRANSFROM
SAMPLE DATA
CREATE
SAMPLE DATAS
LIMITS
SHEAR RANGE
ZOOM RANGE
WIDTH_SHIFT
HORIZONTAL _FLIP
ROTATION_RANGE
HEIGHT_SHIFT
FILL_MODE
FILL
NEW PIXELS
ROTATION
WIDTH
SHIFT
HEIGHT
REMIX
OLD INFORMATION
