Import all the models from Caffe ModelZoo into IDE

[dexter1691]

We should try to successfully import all the models from Caffe ModelZoo to see if some layers are not working. Consequently fix or add the missing the layers into the IDE.

https://github.com/BVLC/caffe/wiki/Model-Zoo

For GSOC participants: This seems like a very good starting point to see how IDE works. Here is what you have to do

• Choose one of the models from the link above and create an issue for it. Cross reference it here for others to see.
• We will assign the issue to you
• If the model is working. Add a unit test for it so that any later changes to the projects ensure compatibility with those models.
• If the model isn't working, try to discuss with the mentors and fix the issue!

We are maintaining the list of models tested by contributors here: https://github.com/Cloud-CV/IDE/wiki/Models

[muddlebee]

https://gist.github.com/jimmie33/7ea9f8ac0da259866b854460f4526034

This is working for me. @dexter1691

Edit : Ohh I just checked this has already been tested. present in examples/ folder.

[dexter1691]

Nice! Let's maintain a list of the one's that working and the ones that are not working. GoogleNet was already part of the ones that we used to check.

Let's keep trying others to find out the ones that are not working. I am sure there will be a few!

[muddlebee]

https://gist.github.com/jimgoo/0179e52305ca768a601f Not working.

Edit: This should work. All models are present itself.

[kris-singh]

I would like to work on this

[virajprabhu]

@anweshknayak Are you sure we have front-end support for all layers in the prototxt?

@kris-singh Great, can you start with simple classification models like AlexNet and VGGNet?

Would be great if either of you could try out a ResNet (it won't work at the moment) and open an issue with the layers that we are missing. This task would also give you a good overall picture of this project's current status.

Edit: I see @dexter1691 has already opened an issue for Resnets in #31. Could we confirm that the layers he listed are the only ones we are missing front-end support for?

[Ram81]

@virajprabhu I am interested in this one can you guide me?

[virajprabhu]

@Ram81 Great. Have you set up IDE and tried importing the example prototxt file for GoogLenet? We similarly want to be able to import a wide variety of prototxt files from the Caffe Model Zoo. For that, we may need to add front-end support for more Caffe layers. As a concrete starting point, can you try importing classification CNN models like AlexNet / VGGNet / ResNet and verify that they work (by commenting on this thread) / identify which layers are missing and open an issue for the same?

[Ram81]

@virajprabhu sure I will setup IDE and let you know if I need some help :+1:

[utsavgarg]

@virajprabhu I am also interested in trying out classification networks if any hasn't been implemented yet

[utsavgarg]

@dexter1691 AlexNet from Caffe Model Zoo loads fine, but the 'Group' parameters present in conv layers are ignored. https://github.com/BVLC/caffe/blob/master/models/bvlc_alexnet/deploy.prototxt VGG-16 present in Caffe Model Zoo is for an older Caffe version, I updated the prototxt to support the newer Caffe version and it loads fine now. https://gist.github.com/utsavgarg/e3982498e9fb27d882cf877e7992ec9b

[Omarito2412]

Squeezenet

[Omarito2412]

RCNN

[utsavgarg]

Network in Network

[virajprabhu]

@utsavgarg @Omarito2412 Thanks for reporting these, and opening issues for the ones that don't work! If you would like to work on adding support to any of the ones that don't work, post on the issue thread and I'll assign it to you. Feel free to ping the Gitter channel / comment if you need any help setting up or would like to discuss the implementation.

[Omarito2412]

@virajprabhu I would like to work on adding support for S2VT network #36

[virajprabhu]

@Omarito2412 Sounds good. Please go ahead, and let me know if you need help.

[Ram81]

@virajprabhu which one should I try can you suggest one?

[prasadwajekar]

I would like to work on this...

[AhmedAbdulKarim]

how to be assigned to this?

[virajprabhu]

@prasadwajekar @AhmedAbdulKarim You can try importing popular models from the Caffe Model Zoo, and comment on this thread with the status (success/fail). If it fails and you'd like to add support for the missing layer(s) required to get it to work, please submit a PR!

[Ram81]

@virajprabhu @utsavgarg "ResNets: Deep Residual Networks from MSRA at ImageNet and COCO 2015" doesn't currently get loaded in fabrik can I add support for it?

[utsavgarg]

@Ram81 ResNet-101 is already there in the Fabrik Zoo and it loads fine. Which model were you trying ?

[Ram81]

@utsavgarg I tried ResNet-101-deploy.prototxt from https://github.com/KaimingHe/deep-residual-networks/tree/master/prototxt

[utsavgarg]

@ram81 you will need to change the format of the input layer in that prototxt.

[Ram81]

Okay understood I'll try and find which one doesn't work & ask further

[Ram81]

@utsavgarg should I also report which one already works so that we can add it to Fabrik/Zoo?

[JPJPJPOPOP]

These two don't work due to being in the old Caffe format: https://github.com/RankingCNN/Using-Ranking-CNN-for-Age-Estimation/blob/master/basic_16to66/deploy_basic.prototxt https://gist.github.com/mavenlin/d802a5849de39225bcc6#file-train_val-prototxt

Will try updating them to see if they work.

[JPJPJPOPOP]

Finished updating both: RankingCNN: https://gist.github.com/JPJPJPOPOP/6ed94b737a45da81fbc55ce912d6da2a

NINImageNet: https://gist.github.com/JPJPJPOPOP/b8d9f5a1c4cd1bbe624f664f1b85d9bb