Albumentations Tutorial for Data Augmentation (Pytorch focused)

Albumentations is the way to go. I really like this library and I think you will too!

❤️ Support the channel ❤️

Paid Courses I recommend for learning (affiliate links, no extra cost for you):

✨ Free Resources that are great:

💻 My Deep Learning Setup and Recording Setup:

GitHub Repository:

✅ One-Time Donations:

▶️ You Can Connect with me on:

Image stolen from this awesome article:

Timestamps:
0:00 - Introduction
1:21 - Augmentation for Classification
9:11 - Augmentation for Segmentation
11:55 - Augmentation for Detection
21:35 - Full PyTorch Example
31:05 - Ending

[AladdinPersson]

A mistake I made in the video was for the full pytorch example when creating the dataset I accidently inherited from nn.Module instead of Dataset. We didn't get an error because we didn't create a DataLoader, will fix that for the code I upload to Github 🤙

[shahidsiddiqui]

Man your tutorials are better than a bunch of paid courses out there :) Thank you so much.

[billfuture]

Great video for a postgraduate student worked in deep learning!!!I've followed and introduce to other mates.

[LocNguyen-lbii]

OMG your tutorial helps saving a ton of hour preprocessing.

[gilpalmafernandes]

Excellent and concise tutorial on Albumentations. Many thanks!

[nmpwl]

These are excellent tutorials, thank you very much! 😃

[adhoc]

Loving the videos. Keep uploading more, please!

[guyindisguise]

Great video, could you make one about implementing GradCAM (and/or other interpretability tools) for popular models (like resnet) in pytorch?
Would appreciate it very much.

[zeyutang]

Great Video! I wonder how are we gonna resize the predicted bboxes of the test set back to its original size

[thepresistence]

Hey, I have doubt, when we are using this in our custom class, Before and after augmentation size of the data is same.


train_data = 300 (before augumentation)
train_data = 300 (after augumentation)

There is no increase in size of the data, if we use use for loop, it's creating more data but when we are doing in custom class, it's just changing the data not increasing the train_data, can you please help with that.

[ranns]

Great video, thank you !!
Also, how do I use this on standard datasets like CIFAR? I assume we cannot use these albumentations inside the "transform" argument of torchvision.datasets.Cifar10?

[Kwaiigurlie]

Thank you Aladdin for excellent explanation. I have one query that the image size and mask size of elon image are not same. Therefore, the segmentation script gives an error "ValueError: Height and Width of image, mask or masks should be equal." please help in this

[QuarktaschemitSenf]

Hi there. First of all ty for the Video. But i have a question: Currently i am getting into Object Detection with yolov7. For each image i have a txt-file with the coordinates of the bounding boxes in yolo format. How do i apply the rotation or for example a flip to the bounding boxes? Can you maybe do a video on that?

[kae]

Loving the videos! Do you plan on continuing with the App Deployment Series?

[sahil-]

Great tutorial!!😁
I am looking for efficientNet model from scratch ad well as review of this research paper on upcoming videos.

[smoothumut]

great job friend. A Quick question, you didnt increase the number of dataset by applying transform, did you?

[Joseph-ubrh]

I'm using your deeplearning conda environment but cannot seem to get albumentations installed.

[vijayvikasmangena]

module 'albumentations' has no attribute 'BboxParams'

Very strange.What am I missing here?I have tried uninstalling and re install the package.But it didn't work.

[xtraeone]

When we augment the images what happens to our annotations files? Shouldn't the respective annotations be made for every new augmented image ?

[md.shamimalmahmud]

How can I use the dataset which you have used in this video?