I made available VGG16, VGG19, ResNet50: Keras code +ImageNet weights for both TF and Theano.https://github.com/fchollet/deep-learning-models …
-
-
Replying to @fchollet
You can now literally do: ``` model = VGG16(weights='imagenet') preds = model.predict(imgs) print(decode_predictions(preds)) ```
6 replies 45 retweets 152 likes -
Replying to @fchollet
Hoping to see great new models from the community --NLP, vision, speech, and more.
1 reply 2 retweets 12 likes -
Replying to @fchollet
We have a complete ASR workflow in Torch, would love to port it over to Keras, just need some good CNN>LSTM examples to build on.
1 reply 0 retweets 1 like -
Replying to @bentaylordata
we actually do have a number of CNN + LSTM examples in the main repo. Looking for something specific?
1 reply 1 retweet 1 like -
Replying to @fchollet
audio > spectrogram (2d matrix, y axis is frequency, x is time), 2Dmat > convnet, then slide it by time.pic.twitter.com/QqmCk0PwQu
1 reply 0 retweets 1 like -
Replying to @bentaylordata @fchollet
the raw output of the conv net needs to be passed through the LSTM by time. Ideal for speech.https://arxiv.org/abs/1512.02595
4 replies 0 retweets 1 like
so a stack of Conv1D and MaxPooling1D followed by LSTM https://github.com/fchollet/keras/blob/master/examples/imdb_cnn_lstm.py#L58-L64 …
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.