Build a Deep Learning Project From Scratch | Hands-on Tutorial

Share it with your friends Like

Thanks! Share it with your friends!

Close

🙋‍♂️ We’re launching an exclusive part-time career-oriented certification program called the Zero to Data Science Bootcamp with a limited batch of participants. Learn more and enroll here: https://www.jovian.ai/zero-to-data-science-bootcamp

🔗 Resources used
• Notebook created in the workshop: https://jovian.ai/aakashns-6l3/deep-learning-project-live
• Guidelines and datasets for deep learning projects: https://jovian.ai/learn/deep-learning-with-pytorch-zero-to-gans/assignment/course-project

💻 In this live hands-on workshop, we’ll build a deep learning project from scratch in 2.5 – 3 hours. You can follow along to build your own project. Take our Free Certification Course “Deep Learning with PyTorch: Zero to GANs” to learn the required skills: http://zerotogans.com

Here’s an outline of the workshop:
📄 Find an interesting unstructured dataset online (images, text, audio, etc.)
❓ Identify the type of problem: regression, classification, generative modeling, etc.
🤔 Identify the type of neural network you need: fully connected, convolutional, recurrent, etc.
🛠 Prepare the dataset for training (set up batches, apply augmentations & transforms)
🔃 Define a network architecture and set up a training loop
⚡ Train the model and evaluate its performance using a validation/test set
🧪 Experiment with different network architectures, hyperparameters & regularization techniques
📰 Document and publish your work in a Jupyter notebook or blog post

📒 Datasets from the workshop:
Chest X-Ray Images (Pneumonia) – https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia
Fruits 360 – https://www.kaggle.com/moltean/fruits
Flowers Recognition – https://www.kaggle.com/alxmamaev/flowers-recognition
Malaria Cell Images Dataset – https://www.kaggle.com/iarunava/cell-images-for-detecting-malaria
Intel Image Classification – https://www.kaggle.com/puneet6060/intel-image-classification
Best Artworks of All Time – https://www.kaggle.com/ikarus777/best-artworks-of-all-time
CelebFaces Attributes (CelebA) Dataset – https://www.kaggle.com/jessicali9530/celeba-dataset

Open Datasets – https://github.com/JovianML/opendatasets

⚙ Check out these projects for inspiration:
• Blindness Detection using Image Classification: https://medium.com/jovianml/blindness-detection-using-image-classification-5d86d9e47870
• Generating New Artworks using GANs: https://medium.com/jovianml/generating-art-with-gans-352ceef3d51f
• Bounding Box Prediction using PyTorch: https://towardsdatascience.com/bounding-box-prediction-from-scratch-using-pytorch-a8525da51ddc
• Classifying Environment Audio Recordings: https://medium.com/jovianml/classifying-environmental-audio-recordings-9500a2112234

👨‍🏫 This workshop is taught by Aakash N S who is the co-founder and CEO of Jovian – a community learning platform for data science & ML. Previously, Aakash has worked as a software engineer (APIs & Data Platforms) at Twitter in Ireland & San Francisco and graduated from the Indian Institute of Technology, Bombay. He’s also an avid blogger, open-source contributor, and online educator.

#DeepLearning #Project #PyTorch #Walkthrough #DataScience
#CNN #ImageClassification #GANs #MachineLearning


Learn Data Science the right way at https://www.jovian.ai
Interact with a global community of like-minded learners https://jovian.ai/forum/
Get the latest news and updates on Machine Learning at https://twitter.com/jovianml
Connect with us professionally on https://linkedin.com/company/jovianml
Follow us on Instagram at https://www.instagram.com/jovian.ml/
Join our Telegram community at https://t.me/jovianml
Subscribe for new videos on Artificial Intelligence https://youtube.com/jovianml

Comments

Murtaza Pakawala says:

amazing lecture

Alptug Aydin says:

How do we decide the size after applying Pooling? I mean the size is halved in 1:29:00 after MaxPool but what decides to halve it or what is the formulation to calculate the new size?

Ahamed Basha N says:

There is no one in this world who can teach each and everything so clearly and descriptive like you. Hats-off sir.

CIVILTOCODER says:

How to predict image with help of .pth model in new Jupiter notebook can anyone help

YTG says:

Sir, 1:00:00 ke baad se samaj nhi aaya 😭😭😭😭

Aniket Kumar says:

Get your hands dirty on project. Stop following tutorial

Aflah says:

Unfortunately I missed the stream but I did do the ZerotoPandas course and was so proud of my final project. I've now started the ZerotoGANS series and hope to learn many more skills from this wonderful channel.

jonathyn blythe says:

Wow, thank you!

Ulises Salvador Gomez says:

I hope they do something like ZeroToGANs but instead be ZeroToTransformers. That would be amazing!

sat k says:

Fantastic!!! I managed to complete a project although I was scared to 🙂 Thank you 🙏

nasiru sule says:

After completing all assignments in the course, i had already given up in carrying out the course project until i came across this video. Thank you Mr Aakhash for this wonderful video, you are a fantastic teacher.

Annan says:

Hello Aakash,

I was looking at the Kaggle competition launched by Jovian a couple of months ago and I was wondering how one can get access to the dataset to play with it to learn more about multilabel classification as it says, "This is a limited-participation competition. Only invited users may participate." and when trying to download the dataset, "Unable to perform operation since you're not a participant of this limited competition." So we can't view the dataset, please let us know how we can access it,

Thanks!

Daniel Krüger says:

Hi there, I just tested AdaptiveMaxPool2d function, in order to be able to use my trained model with any picture size, unfortunately the pictures are only recognized correctly, if they are in the same size as the training pics (32×32). Thats my model, any hints how to get it to work?

Epoch [29], train_loss: 0.0483, val_loss: 0.2013, val_acc: 0.9412

class Letters(ImageClassificationBase):
def __init__(self):
super().__init__()
self.network = nn.Sequential(
#nn.AdaptiveAvgPool2d(output_size=32),
nn.Conv2d(3, 32, kernel_size=3, padding=1),
nn.ReLU(),
nn.Conv2d(32, 64, kernel_size=3, stride=1, padding=1),
nn.ReLU(),
nn.MaxPool2d(2, 2), # output: 64 x 16 x 16

nn.Conv2d(64, 128, kernel_size=3, stride=1, padding=1),
nn.ReLU(),
nn.Conv2d(128, 128, kernel_size=3, stride=1, padding=1),
nn.ReLU(),
nn.MaxPool2d(2, 2), # output: 128 x 8 x 8

nn.Conv2d(128, 256, kernel_size=3, stride=1, padding=1),
nn.ReLU(),
nn.Conv2d(256, 256, kernel_size=3, stride=1, padding=1),
nn.ReLU(),
#nn.MaxPool2d(2, 2), # output: 256 x 4 x 4
nn.AdaptiveMaxPool2d(1),

nn.Flatten(),
nn.Linear(256, 1024),
nn.ReLU(),
nn.Linear(1024, 512),
nn.ReLU(),
nn.Linear(512, 33))

def forward(self, xb):
return self.network(xb)

Daniel Krüger says:

I followed the Zero to GANs course because I wanted to learn how to train models by myself. This course is just amazing and super interesting. This video as an addition to the course makes it more clear to understand the big picture of ML. Big thank you for this amazing work 👍

diego gachet says:

this notebook is available ?

San dro says:

Hope you are not taking it hard on yourself for students falling behind original plans. First is that many has fulltime jobs, family, getting sick etc that keeps them occupied. Also we are so used to being overstimulated these days with everything from games, youtube, porno (very stimulating) that we are not used to thinking these days.

Rodo says:

It's so motivating to watch your workshop video. Even just the playback from live. <3

Karthik Jangid says:

Jovian is the future of ML/DL education

Akanksh Adugani Vanjari says:

Really loved your video!!!
The best thing you have done is you have sent the invitation to all the colleges which made me land over here!
I guess now your main target should be advertising it, because not many are sure of the company or channel, but I think this is one of the best channels in ML/DL the content you have is crisp and clear once people get to notice your work then I think they will get in love with your videos and this i how i am feeling. I am from IIT Tirupati, a small request from my side can you frequently send invitations to our institution regarding all your new updates and live streams and all the other stuff which you want to share we glad we got to know about your company. and I thoroughly enjoyed watching your video. now I am also going to watch all your videos on zero to gans i feel like those videos will take my knowledge to another level in this field.

Also is there anyway apart from youtube wherein I can get to subscribe to all your works and new updates?

A small suggestion if you say that you are from IITB then I think there will be more viewers because people always want to learn stuff from top IITans like you.

And i wish you all the very best for helping us in getting to know more about DL and keep posting such interesting videos .

Write a comment

*

Area 51
Ringing

Answer