Comments (22)
I converted the VGG-16 caffemodel to TensorFlow https://github.com/ry/tensorflow-vgg16 which might be helpful for a generalized script
from tensorflow.
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.
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.
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.
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.
+1
from tensorflow.
👍
from tensorflow.
Reopening to de-dupe future requests for this
from tensorflow.
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.
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.
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.
@shlens, can you comment on this?
from tensorflow.
@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.
for this inceptiion5h model, anyway to see the code that it was trained with?
from tensorflow.
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.
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.
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.
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.
pls check answers and comments on stackoverflow link @ProGamerGov
from tensorflow.
@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.
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.
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)
- Tensorflow TF ops - CPU Factory not registered HOT 8
- Complex dtype input for keras layer in tf2.16+
- [Android x86_64] [TFLite] `Fatal signal 11 (SIGSEGV), code 128 (SI_KERNEL), fault addr 0x0` crash from `libtensorflowlite.so` HOT 1
- [DOCS] Missing complex input for Round op HOT 1
- Ensemble provides either correct or (the same) incorrect inference results HOT 1
- tf.keras.utils.plot_model doesn't work
- tf.raw_ops.Unbatch aborts with "Check failed: d < dims()" HOT 3
- Incorrect documentation of the hard_sigmoid activation function (TensorFlow 2.16.1) HOT 4
- CategoryEncoding layer with output_mode count as part of a model no longer working in TensorFlow 2.16.1 HOT 1
- TimeDistributed layer with nested model no longer working in TensorFlow 2.16.1
- Loading of a model with a Bidirectional LSTM layer no longer working in TensorFlow 2.16.1
- Documentation for RaggedTensors does not match Input layer documentation
- Tensorflow doesn't work on my data HOT 2
- Segmentation fault when using tflite_model_maker searcher.TextDataLoader.create(EmbeddingModel, l2_normalize=True)
- ValueError: Only instances of `keras.Layer` can be added to a Sequential model. " Keras version - 3.5.0" HOT 2
- model.summary() broken for custom models subclassed from tf.keras.Model
- opal_wrapper-wrapper-data.txt not found HOT 2
- Unable to run Object Detection API i.e. AttributeError: module 'keras._tf_keras.keras.layers' has no attribute 'experimental' HOT 1
- loss growing to negative infinity val accuracy remaining constant, testing accuracy above 100%
- TFLite Op type not registered (RegexSplitWithOffsets) in Swift
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
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.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from tensorflow.