AI Computer Vision

intro

• motivation Exploiting Data Structure

• cnn example for Frequency Estimation

  • lab Convolution for Frequency Estimation 📔show also when model dont learn to generalize

• Basic convulsion padding and basic kernel examples

• 2d convolution

basic features

• layers

• pooling

• batch Norm

  • Norm demo

• filters

  • laplacian

pytorch layers

• common layers

• custom nn

• MulticlassAccuracy micro macro modes

lib

• torchvision lib

• v1/v2.transforms

concepts

• receptive field

  • lab calc RF📔

• encode decode

• skip connections

augmentation

• about augmentation

• basic img effects📔

• cut effects

  • lab demo cut effects📔

• smoothing

  • lab smoothing demo📔

• lab combine cutting with smooth📔

demo arch

• lab cifar📔

• lab mnist 📔- multiple models with tensor board

• lab student model 📔- Knowledge Distillation

common arch

• compare

• darknet53

• alexnet

• vgg

• googleNet

  • Inception Modules

• resnet

  • lab resnet model📔

• yolo

• minirocket

  • lab minirocket demo📔

inference

• preDefine models

• pytorch inference

• sota

• zoo

• lab pre train compare📔

transfer learning

• about transfer learning

• compare methods

• fineTune

• freeze

• lab Transfer Learning of resnet📔

• overfit

  • lab reduce demo📔 TBD

• transformers

  • lab transofmers basic📔 TBD

• adaptation

debug by visualize

• methods

location

(object) location
    
    localization                - single object in the image
        classification          - what the object
        regression              - where the object
    
    detection                   -no knowledge of how many objects are in the image
        Segmentation
            - classification    - what the object
            - pixel regression  - where the object

• about location

  • lab pretrain📔

localization

• localization

  • lab localization📔

detection

• detection

  • lab pretrain detection demo📔

• common arch

• NMS Non Maximum Suppression

• MAP

• workshop object Detection with Yolov8

  • labelme

  • 0001BoundingBoxFormat

  • 0002Dataset

  • 0003PreProcessTiles

  • 0004PreProcessTrainTestSplit

  • 0005TrainYolo

Segmentation

• about Segmentation

  • intro

  • data

  • models

  • unet

  • workshop segmentation UNET

    • 0001Dataset

    • 0002TrainModel

    • 0003InferModel