EfficientNet! - Keras Code Examples

[@oussamayousre7845]

guess what ! i just discovered a Great channel , keep that amazing work

[@hoaxuan7074]

The central limit theorm (CLT) applies not just to sums but equally to sums and differences. The fast Hadamard transform is done using patterns of addition and subtraction, Hence the CLT applies. Therefore transforming a vector of random numbers drawn from the uniform distribution results in a vector of random numbers from the Gaussian distribution. One technical issue is the transform leaves vector length unchanged. And the Gaussians are ever so slightly entangled as a result. Still it is an extremely fast generation method. There is a 1969 paper about it. There are multiple reasons everyone should study the fast Hadamard transform again after a 50 year gap, if you ask me.

[@hoaxuan7074]

And I wasn't asked. 😆😆😆

[@radhikapatil8003]

Please do custom image classification using efficientnet..

[@LAKXx]

Very helpful ! thank you good sir

[@rishiktiwari]

At 15:35, don’t we need to set model.trainable = 0 and then partially freeze/unfreeze layers? I mean the note that is written above the code snippet.

[@oussamayousre7845]

please i have a question , when you defined the EfficientNetB0 from scratch , i think you used the baseline model mentioned in the paper , but when we already train a model and save weights and we just want to increase the accuracy , am i able to just feed it(the EfficientNet Model) my own weights and only change the top layer and re-train the model , like is it able to scaling up my own model using only loaded weights ? , or i just need to use the pre-trained weights , ImageNet for example , configure the EfficientNet model and make suitable for my Dataset then train the model !!

[@connor-shorten]

Hey Oussama, you can load in your own weights with the same save and load interfaces, you have the option include_top=False if you want to attach a new output layer -- you can put it in the Sequential model API to do this, for the ImageNet weights there is a "weights argument" -- check out this documentation: keras.io/api/applications/efficientnet/. Hope that helped!

[@아이고어쩌면좋누]

Awsome thanks !!!

[@Rivhoy6ivu7r]

Really helpful. Thanks.

[@connor-shorten]

Thank you so much!

[@ericbly3517]

UnimplementedError: File system scheme '[local]' not implemented (file: '~/tensorflow_datasets/stanford_dogs/0.2.0/stanford_dogs-train.tfrecord-00000-of-00004')

[@teddy911]

where is the code link