Icevision: Tutorials and Examples

[lgvaz]

📒 Tutorials

Tutorials are in .ipynb format, explaining each step of the process, really detailed, not production like.

Core

• [x] [parser](https://lgvaz.github.io/mantisshrimp/tutorials/parser.html)

Object detection

• [x] [wheat](https://lgvaz.github.io/mantisshrimp/tutorials/wheat.html)
• [x] [wheat with detr](https://lgvaz.github.io/mantisshrimp/tutorials/hub_detr_finetune_wheat.html)

Segmentation

• [ ] [penfundan](https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html)

Keypoints

📓 Examples

Examples are be in the .py format, more production oriented. Ready to be run with arguments from the command line and easy to integrate with wandb sweeps and alike.

Object detection

• [ ] [wheat](https://www.kaggle.com/c/global-wheat-detection)

Segmentation

Keypoints

---

Is there a new tutorial or example you would like to add? Comment below and we talk about it 😁

Once we agree, create an Tutorial or Example request issue (use the template) and I'll edit this post with your new cool example!

[lgvaz]

It would be a cool addition to show how to do a wandb sweep in the wheat example

[chho-work]

Certainly it would be! Wandb supports Pytorch Lightning: https://docs.wandb.com/library/frameworks/pytorch/lightning

I will work on few examples!

[lgvaz]

It might be a good idea to separate this into Examples and Tutorials.

Tutorials would be in .ipynb format, explaining each step of the process, really detailed, not production like.

Examples would be in the .py format, more production oriented. Ready to be run with arguments from the command line and easy to integrate with wandb sweeps and alike.

What do you think @chho-work ?

[chho-work]

Yes, it would be great to have both available, examples and tutorials, each with their corresponding formats.

[chho-work]

I found this pretty interesting! I think we can use them in tutorials and examples.
https://github.com/mseg-dataset/mseg-api

[tazu786]

Hi, I have question regarding the new colab on detr fine tuning. Right after loading the checkpoint with no head, shouldn't I freeze all the other layers with something like:

for param in model_without_ddp.parameters():
param.requires_grad = False
model.class_embed.weight.requires_grad = True
model.class_embed.bias.requires_grad = True

Thanks

[lgvaz]

That's something we could do as well correct, we actually have an issue for that in #72

[lgvaz]

We should change the definitions of tutorials and examples.

Tutorials should be non-repetitive, and should all show in the documentation. Having too many tutorials might confuse beginners! Tutorials should be maintained and created by the core team.

Examples should be a mix of core/community, the can be scripts or notebooks and don't need to explain library concepts. There is no limit to the number of examples we should have, the more the better!

Examples also go very well together with datasets implementations in hub. Maybe all examples should actually be all inside hub? This would make them always self-contained (all functions for downloading data, creating parsers, etc would be present)

@ai-fast-track what do you think?