Blog posts from 2 weeks ago
The group is ending?
This is the blog post on Optimizers - “Adam” and friends | Committed towards better future
Edit: Aman mentioned it too. I spoke too soon
We also have a PyTorch study group too where the PyTorch topics will be covered soon for anyone interested
Hopefully others will have a better idea, but my “big project” from that break between fastai versions, which took over my life was Chai Time Data Science
Thanks Aman for doing these sessions. You have given us enough momentum to continue the learnings. Looking forward to stay connected in the Paper Reading Group.
How to get source code in vscode?
I have a question on Head. For Classification, we could use certain architectures (VGG, ResNet etc). But for Segmentation or Object Detection, we usually hv different base architectures like UNet / RetinaNet etc.
Most of the work is really in the Data Prep. Can you speak a bit about Data Prep step in Object Detection. In it, we have location of objects, but the architecture might use Anchors or other techniques. So we no longer have the 1:1 we used to have in Classification or even Segmentation. How do you go about building Dataset for Object Detection tasks where each architecture might have different losses / outputs (unlike Classification or Segmentation tasks).
One neat trick you can try-> Head over to fastai repo on GH. And press dot on your keyboard
(It launches a VSCode instance, not powerful enough for DL but p great for reading source code)
If that doesn’t work change github to github1s
We ran a hugging face study group too
NLP course by Rachel - new fast.ai course: A Code-First Introduction to Natural Language Processing · fast.ai
Maybe I could start a fastai (NLP) study group next
Once you get to transformers - here is another resource to help with your journeys! The Annotated GPT-2 | Committed towards better future
Professor Zach had created one
Two more resources on CV:
Segmentation models pytorch is another great repository - GitHub - qubvel/segmentation_models.pytorch: Segmentation models with pretrained backbones. PyTorch.
This will always be remembered as our favourite (unofficial) fastai 2021 course. Thank you for hosting it!