Giter VIP home page Giter VIP logo

Comments (22)

ry avatar ry commented on March 29, 2024 13

I converted the VGG-16 caffemodel to TensorFlow https://github.com/ry/tensorflow-vgg16 which might be helpful for a generalized script

from tensorflow.

pkmital avatar pkmital commented on March 29, 2024 6

As part of our free course on Creative Applications of Deep Learning, I've included code for loading pre-trained networks for inception v3,v5, vgg16, vgg face, i2v, and i2v tag: https://github.com/pkmital/CADL/tree/master/session-4/libs

from tensorflow.

AshishBora avatar AshishBora commented on March 29, 2024 3

The link to pretrained model of inception for imagenet seems broken
https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/android/assets/tensorflow_inception_graph.pb

from tensorflow.

petewarden avatar petewarden commented on March 29, 2024 2

You can find one pre-trained network as part of the Android example, at https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/android/assets/tensorflow_inception_graph.pb
This implements a version of the Inception architecture for Imagenet classification.

from tensorflow.

shlens avatar shlens commented on March 29, 2024 1

Yes, Inception is known to be far smaller in terms of parameter count then VGG. The first version of Inception contained roughly 5M model parameters; VGG was >75M model parameters (I don't know the exact the number for VGG). As @vodp mentions, the primary reason for this difference in parameter count is due to the fact that VGG uses several large fully connected layers at the top where as Inception minimizes to just 1 layer for classification.

from tensorflow.

raingo avatar raingo commented on March 29, 2024

+1

from tensorflow.

sjuvekar avatar sjuvekar commented on March 29, 2024

👍

from tensorflow.

vrv avatar vrv commented on March 29, 2024

Reopening to de-dupe future requests for this

from tensorflow.

petewarden avatar petewarden commented on March 29, 2024

This is working as expected. We've moved the model out of the source repo, but it's available for separate download. See the android example README for the latest instructions:
https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/android/README.md
Currently the link is https://storage.googleapis.com/download.tensorflow.org/models/inception5h.zip but the README will hold the most up-to-date location going forward.

from tensorflow.

martinwicke avatar martinwicke commented on March 29, 2024

The inception model checkpoint is downloadable (and we have examples using it for classification and deepdream), and more models are being added to https://github.com/tensorflow/models. Closing this.

from tensorflow.

vodp avatar vodp commented on March 29, 2024

I found that the file size of https://storage.googleapis.com/download.tensorflow.org/models/inception5h.zip is very small, ~50MB. The Inception v3 model in tensorflow/tensorflow/examples/label_image/data/inception_dec_2015.zip is rather larger ~100MB but still portable on mobile devices. However when I tried to convert VGG-16 models into .pb file, it cost ~500MB. Is that the architecture of Inception is more superior than VGG in term of portability?

from tensorflow.

martinwicke avatar martinwicke commented on March 29, 2024

@shlens, can you comment on this?

from tensorflow.

vodp avatar vodp commented on March 29, 2024

@martinwicke I think my guess is right. I tried to freeze the inception-v1 (GoogleNet) and found it just 27MB, a half of inception-v3. This is probably due to the less number of fully connected layers used (infact just one layer in the case of inception-v1) compared to VGG-x architectures. This turns out that inception arch is really friendly for embedding apps.

from tensorflow.

jtoy avatar jtoy commented on March 29, 2024

for this inceptiion5h model, anyway to see the code that it was trained with?

from tensorflow.

GeorgianaPetria avatar GeorgianaPetria commented on March 29, 2024

Is there any way to load a graph partially in tensorflow?

I would like to be able to use the weights trained for classification, but replace the fully connected layers and retrain for regression.

This can be done in Keras using the "by_name" option when loading a model, but I couldn't find a way to do it in tensorflow.

from tensorflow.

petewarden avatar petewarden commented on March 29, 2024

You might take a look at the TensorFlow for Poets codelab and sample code to see a similar approach to removing the final fully-connected layer:
https://codelabs.developers.google.com/codelabs/tensorflow-for-poets/index.html#0
https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/image_retraining/retrain.py

from tensorflow.

GeorgianaPetria avatar GeorgianaPetria commented on March 29, 2024

Thanks @petewarden!
The code in retrain.py is very helpful!
However, since inception has 3 output layers, don't I need to retrain all the 3 outputs?
In the retrain.py code a new layer is added on top of the bottleneck 'pool_3', but won't my regression be affected by not replacing all the outputs?

from tensorflow.

ProGamerGov avatar ProGamerGov commented on March 29, 2024

The inception5h model doesn't look like it can be used for classification or transfer learning, because it is missing the required layers: https://stackoverflow.com/questions/42003846/retraining-inception5h-model-from-tensorflow-android-camera-demo

Is there an un-stripped version of the model?

from tensorflow.

astrung avatar astrung commented on March 29, 2024

pls check answers and comments on stackoverflow link @ProGamerGov

from tensorflow.

Charles808 avatar Charles808 commented on March 29, 2024

@GeorgianaPetria Hi ! I also want to add regression to the current inception v3 model.
if you already did it. Do you have any useful clue for it ?

Thank you b4.

from tensorflow.

rish321 avatar rish321 commented on March 29, 2024

Hi guys, is there any pre trained model file for person identification in tf using inception? Re-training isn't working for me and I need to use model file in my Android project.

from tensorflow.

mrgloom avatar mrgloom commented on March 29, 2024

How to see content of http://download.tensorflow.org/models/ ? i.e. I want to see full list of models that are available.

from tensorflow.

Related Issues (20)

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.