@4:35: Are you sure that augmentation is done at __getitem__?? To me it seems that augmentation must be done on the whole dataset before training, and it increases the number of images in dataset. Hence it should be a separate step and shouldnt be a part of __getitem__, whose task is to read a single item from the dataset..
Would it be better practice to do len(self.targets) since in a lot of cases, its just 1 image folder path and then you construct the exact image path for a given instance within __getitem__ itself? While targets is usually from a csv
Hi, thank you for your videos! I have a questoin about the `np.transpose function`. as i understand we transform the matrix of rows with pixeld into the separate trices of the separate channels per pixel... if it make sense.
Hey Abhishek, can you suggest me how I should go about learning NLP in a structured way?, I already know the basics, but I wanted to go back to the drawing board and build a better foundation
great work, where is the albumentations video?
Hi, thanks for the video. Great job explaining how this works.
It's awesome 🎉🎉
@4:35: Are you sure that augmentation is done at __getitem__?? To me it seems that augmentation must be done on the whole dataset before training, and it increases the number of images in dataset. Hence it should be a separate step and shouldnt be a part of __getitem__, whose task is to read a single item from the dataset..
thank you. but it is possible to augment an image in more than one way (reflection, playing with colors etc). which augmentation does this program do?
Quality 🔥🔥🔥
great stuff!!
Would it be better practice to do len(self.targets) since in a lot of cases, its just 1 image folder path and then you construct the exact image path for a given instance within __getitem__ itself? While targets is usually from a csv
Hi, thank you for your videos! I have a questoin about the `np.transpose function`. as i understand we transform the matrix of rows with pixeld into the separate trices of the separate channels per pixel... if it make sense.
If we have the images in respective folders for classification how can we pass targets and image paths to the dataset?
you can split on "/" to get the targets, then encode them using LabelEncoder of scikit-learn. everything else remains the same.
Hey Abhishek, can you suggest me how I should go about learning NLP in a structured way?, I already know the basics, but I wanted to go back to the drawing board and build a better foundation