DATASET
SPLIT THE DATASET Train/Test/val
DATA TRANSFORMS AND DATA LOADER
CREATE A PRE-TRAINED VISION TRANSFORMER MODEL
REMOVE CLASSIFIER HEAD OR LAYER
FEATURE EMBEDDING LAYER FOR A BATCH OR ENTIRE
BUILD HYBRID MODEL(VIT FEATURES+CNN PROCESSED FEATURES)
TRAIN HYBRID MODEL
EVALUATE HYBRID MODEL TO FIND BEST MODEL
PREDICTION TAKES PLACE WITH BEST HYBRID MODEL
ONLY IMAGE---CORRECT PREDICTION
GRADCAM --WRONG PREDICTION
by nivi