<3 Thanks for adding an 8th lesson! Much appreciated!
3 Likes
What are the advantages of having square images vs rectangular ones?
4 Likes
You probably have as many talls images as you have wide images, so resizing to a square is kind of the mean point.
5 Likes
Why are images usually squished to squares? I would think that the expected aspect ratio would be best
When you use the dls.new method, what can and cannot be changed? Is it just the transforms?
6 Likes
How to ensure that the feature is not lost during crop/resize?
2 Likes
Just item_tfms and batch_tfms yes, here is the signautre:
def new(self, item_tfms=None, batch_tfms=None):
6 Likes
In general you have to look at a lot of images if you want to make sure nothing important is accidentally cropped out. Typically I’ll use .show_batch() and investigate a few batches.
4 Likes
Can qw have multiple item_tfms applied to the same dataset and create a larger dataset. How do you concatenate multiple item_tfms and does the order matter ?
How can we visualize the images after we perform the data augmentation in the batches?
if I start with 100 images, will it add more images? How can I visualize all these new images.
item_tfms do not create a bigger dataset. If you pass multiple ones, they are composed. You still have the same number of images in your dataset.
1 Like
dls.show_batch() will always show you the augmented version.
4 Likes
Why item_tfms and batch_tfms why not just batch_tfms ? Since all transforms are on the individual images.Or am i missing something?
Thanks, how many more pictures will add to our dataloader?
No. As their name indicates, batch_tfms are applied to a batch of images at a time.
We look at the DataBlock source code at Source Code study group and created videos for each session.
https://forums.fast.ai/t/source-code-mid-level-api/65755
This forum post also contains multiple exmaple of DataBlock API including reading four channel images/multilabel audio/5 different single image single label similar to bears and pets examples.
10 Likes
Is the idea of randomResizedCrop that you’ll make multiple copies of the same photo with different parts/versions included all at once? Or are you randomResizing every photo once?
1 Like
None. You still have the same pictures. What a transform might do (if it does data augmentation) is to return a different result each time you ask for the image. But transforms do not add new raw items (aka new filenames).
2 Likes
Can you use methods like RandomResizedCrop to increase the size of your training set?
thank you so much