[WIP] Add new torchvision models #219

oke-aditya · 2021-08-18T18:21:53Z

Adds the following models.

MobileNetv3: - Mobilenet v3 small, Mobilenet v3 large
Mobilenet Segmentation: - lraspp_mobilenet_v3_large
DeepLab Models: - deeplabv3_mobilenet_v3_large deeplabv3_resnet50
FCN: - fcn_resnet50,
MNasNet: - mnasnet0_5, mnasnet1_0
EfficientNet: efficientnet_b0, efficientnet_b1, efficientnet_b2, efficientnet_b3, efficientnet_b4, efficientnet_b5, efficientnet_b6, efficientnet_b7

I modified the DeepLab and FCN files inplace to avoid duplicating. It needed a small rename in the file.
I have added a couple of images denoting architecture of mnasnet and mobilenetv3.
Let me know if those images are fine.
The model description is pending for lr_aspp, mnasnet and MobileNetV3.

But a small review at this point would be great.

cc @datumbox @NicolasHug

netlify · 2021-08-18T18:37:55Z

✔️ Deploy Preview for pytorch-hub-preview ready!

🔨 Explore the source changes: 0f739f5

🔍 Inspect the deploy log: https://app.netlify.com/sites/pytorch-hub-preview/deploys/6133c0f1416fd30007c6179e

😎 Browse the preview: https://deploy-preview-219--pytorch-hub-preview.netlify.app

oke-aditya · 2021-08-18T18:54:02Z

The old pages are still persisting along with the new pages. But I don't rename the files they would convey a wrong meaning.

NicolasHug

Thanks a lot for the PR @oke-aditya , this is very useful

I took a brief look even though this is still WIP, LMK when this is ready for a full round. Considering the big amount of changes I would suggest to include the rest of the models in other PRs, too keep reviews manageable.

Also, unfortunately our CI is completely toasted at the moment: https://app.circleci.com/pipelines/github/pytorch/hub/1244/workflows/a94283ed-e5cc-4c8c-b4cf-9ba78a03900c/jobs/1677 because of all those rate API limit issues. Hopefully it should be fixed in a near future, but sadly it means we'll have to wait until the CI is back again to merge this PR (but we can still do most of the review in the mean time)

pytorch_vision_deeplabv3.md

pytorch_vision_fcn.md

oke-aditya · 2021-08-19T12:30:57Z

Something I found out in log

DEPRECATION NOTICE: Builds using the Trusty build image will fail after September 19, 2021

The build image for this site uses Ubuntu 14.04 Trusty Tahr, which is no longer supported.
All Netlify builds using the Trusty build image will begin failing in the week of September 19.

To avoid service disruption, please select a newer build image at the following link:
https://app.netlify.com/sites/pytorch-hub-preview/settings/deploys#build-image-selection

For more details, visit the build image migration guide:
https://answers.netlify.com/t/end-of-support-for-trusty-build-image-everything-you-need-to-know/39004

I think this should be addressed by pytorch.github.io team.

NicolasHug · 2021-08-19T12:35:57Z

Something I found out in log

Thanks for raising this, we'll notify the relevant team

oke-aditya · 2021-08-19T12:46:37Z

Hmm the open in Colab will not work as of now. I think it would need the notebook in pytorch.gihtub.io repo.
Probably in this folder ?

oke-aditya

Would love your thoughts @NicolasHug and @datumbox

I need to go through the blog and papers to update descriptions of these models.

pytorch_vision_deeplabv3_resnet101.md

pytorch_vision_mobilenet_v3.md

oke-aditya · 2021-08-19T13:52:25Z

pytorch_vision_lraspp.md

+
+Here's a small snippet that plots the predictions, with each color being assigned to each class (see the visualized image on the left).
+
+```python


Not sure if it would be good to use torchvision utils.
Since they are from v0.10+ it might not be backward compatible.

Let me know.

NicolasHug · 2021-08-19T17:16:38Z

Something I found out in log

DEPRECATION NOTICE: Builds using the Trusty build image will fail after September 19, 2021

The build image for this site uses Ubuntu 14.04 Trusty Tahr, which is no longer supported.
All Netlify builds using the Trusty build image will begin failing in the week of September 19.

To avoid service disruption, please select a newer build image at the following link:
https://app.netlify.com/sites/pytorch-hub-preview/settings/deploys#build-image-selection

For more details, visit the build image migration guide:
https://answers.netlify.com/t/end-of-support-for-trusty-build-image-everything-you-need-to-know/39004

I think this should be addressed by pytorch.github.io team.

@malfet @seemethere , @jspisak suggested that one of you guys might be able to help with this :) ?
I checked and unfortunately I don't have the rights to update the ubuntu image myself.

oke-aditya · 2021-08-31T08:15:50Z

I will complete the PR this week mostly. Sorry for the delay.

Edit:
I checked the netlify log
The above Deprecation of Ubuntu Image is still valid.
Well the builds would start failing from September 19. Hopefully someone solves it soon.

datumbox · 2021-08-31T09:28:37Z

@oke-aditya it seems to me you missed* one model.

* Or I sneakily added the new EfficientNet to your list. 😄

oke-aditya · 2021-09-04T18:57:25Z

So the only thing left now is to add descriptions about the models.

Mnasnet
EfficientNet
MobileNetv3
Lrsapp

I think I can finish this hopefully over the weekend.

oke-aditya · 2021-09-04T19:00:22Z

pytorch_vision_efficientnet.md

+
+```python
+import torch
+model = torch.hub.load('pytorch/vision:v0.11.0', 'efficientnet_b0', pretrained=True)


I think that it would be wrong to mention pytorch/vision:v0.10.0 as the model was not added till then.
It will be available in the coming 0.11 release.

Should we merge efficientnet now or after the 0.11 release?
I'm happy either ways, we can keep a separate PR open and merge after the release.

Yeah that's a good point... Good call. Do you want to remove this from the PR and send the addition on a separate? We could leave it unmerged until the release.

Cool! I will remove the efficientnet files from current PR and keep them on separate one.
As you suggested, we can merge it after release 👍

NicolasHug · 2021-09-06T10:06:14Z

@oke-aditya thanks for your work on this, in order to keep the reviewing manageable would you mind creating one PR per model?

oke-aditya · 2021-09-06T11:46:10Z

Ok sure no problem. I will close this PR and re-open new ones which add models one at a time.

oke-aditya added 2 commits August 18, 2021 23:42

add new torchvision models

da3f028

fix small bug

fcd9d64

facebook-github-bot added the cla signed label Aug 18, 2021

NicolasHug reviewed Aug 19, 2021

View reviewed changes

pytorch_vision_deeplabv3.md Outdated Show resolved Hide resolved

pytorch_vision_deeplabv3.md Outdated Show resolved Hide resolved

pytorch_vision_fcn.md Outdated Show resolved Hide resolved

pytorch_vision_fcn.md Outdated Show resolved Hide resolved

rename files to check build

6fdd8a4

and lrsapp output

72cd961

oke-aditya commented Aug 19, 2021

View reviewed changes

Merge branch 'master' of github.com:pytorch/hub into add_models

2173963

Merge branch 'master' of https://github.com/pytorch/hub into add_models

07f00b0

NicolasHug mentioned this pull request Sep 2, 2021

Contribute a model to pytorh_hub #226

Open

oke-aditya added 3 commits September 3, 2021 11:49

update branches to main

f0c65f2

add efficientnet

38976f1

adds efficientnet details

0f739f5

oke-aditya commented Sep 4, 2021

View reviewed changes

oke-aditya closed this Sep 6, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[WIP] Add new torchvision models #219

[WIP] Add new torchvision models #219

oke-aditya commented Aug 18, 2021 •

edited

Loading

netlify bot commented Aug 18, 2021 •

edited

Loading

oke-aditya commented Aug 18, 2021

NicolasHug left a comment

oke-aditya commented Aug 19, 2021

NicolasHug commented Aug 19, 2021

oke-aditya commented Aug 19, 2021

oke-aditya left a comment

oke-aditya Aug 19, 2021

NicolasHug commented Aug 19, 2021

oke-aditya commented Aug 31, 2021 •

edited

Loading

datumbox commented Aug 31, 2021 •

edited

Loading

oke-aditya commented Sep 4, 2021

oke-aditya Sep 4, 2021

datumbox Sep 6, 2021

oke-aditya Sep 6, 2021

NicolasHug commented Sep 6, 2021

oke-aditya commented Sep 6, 2021


		Here's a small snippet that plots the predictions, with each color being assigned to each class (see the visualized image on the left).

		```python

[WIP] Add new torchvision models #219

[WIP] Add new torchvision models #219

Conversation

oke-aditya commented Aug 18, 2021 • edited Loading

netlify bot commented Aug 18, 2021 • edited Loading

oke-aditya commented Aug 18, 2021

NicolasHug left a comment

Choose a reason for hiding this comment

oke-aditya commented Aug 19, 2021

NicolasHug commented Aug 19, 2021

oke-aditya commented Aug 19, 2021

oke-aditya left a comment

Choose a reason for hiding this comment

oke-aditya Aug 19, 2021

Choose a reason for hiding this comment

NicolasHug commented Aug 19, 2021

oke-aditya commented Aug 31, 2021 • edited Loading

datumbox commented Aug 31, 2021 • edited Loading

oke-aditya commented Sep 4, 2021

oke-aditya Sep 4, 2021

Choose a reason for hiding this comment

datumbox Sep 6, 2021

Choose a reason for hiding this comment

oke-aditya Sep 6, 2021

Choose a reason for hiding this comment

NicolasHug commented Sep 6, 2021

oke-aditya commented Sep 6, 2021

oke-aditya commented Aug 18, 2021 •

edited

Loading

netlify bot commented Aug 18, 2021 •

edited

Loading

oke-aditya commented Aug 31, 2021 •

edited

Loading

datumbox commented Aug 31, 2021 •

edited

Loading