Learn practical skills, build real-world projects, and advance your career
Created 3 years ago
!pip install jovian --upgrade --quiet#Image Recognition
'''unique aspect ratio : width to height of the img ratio #perform center crop to keep imp part of img
uniform image size : same size to be fed to nn , fit the img size to feature maps, by down-scaling or up-scaling
mean and perturbed img : to perturb input img, intentionally distort pixel by varying them from mean image, make cnn more robust
cal mean of img by calc the avg pixel value across the training dataset
normalize input img: pixel value btw 0 and 1, subtract mean and divide by std., help pixel to have similar data distribution
dimensionality reduction: rgb has 3 channels, reduce it to grey scale with 1 channel
data augmentation: perturbed img are a form of data augmentations, scaling,rotation, affine transforms, reduces overfittng
'''
'unique aspect ratio : width to height of the img ratio #perform center crop to keep imp part of img\nuniform image size : same size to be fed to nn , fit the img size to feature maps, by down-scaling or up-scaling\nmean and perturbed img : to perturb input img, intentionally distort pixel by varying them from mean image, make cnn more robust\ncal mean of img by calc the avg pixel value across the training dataset\nnormalize input img: pixel value btw 0 and 1, subtract mean and divide by std., help pixel to have similar data distribution\ndimensionality reduction: rgb has 3 channels, reduce it to grey scale with 1 channel\ndata augmentation: perturbed img are a form of data augmentations, scaling,rotation, affine transforms, reduces overfittng\n'# setting up a vm in gcp for dl
# # img preprocessing techniques