Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

[Feature] Provide SavedModels of pretrained models

See original GitHub issue

Can you provide SavedModels of pretrained models (with the saved_model.pbfile)?

Issue Analytics

State:
Created 4 years ago
Reactions:3
Comments:5

Top GitHub Comments

1reaction

shoegazerstellacommented, Dec 23, 2019

@JeffreyCA did you succeed in properly convert the SavedModes into tf.js? I used the method explained here but had these warnings during conversion:

WARNING:tensorflow:Didn't find expected Conv2D or DepthwiseConv2dNative input to 'batch_normalization_8/cond/FusedBatchNorm'
WARNING:tensorflow:Didn't find expected Conv2D or DepthwiseConv2dNative input to 'batch_normalization_20/cond/FusedBatchNorm_1'
WARNING:tensorflow:Didn't find expected Conv2D or DepthwiseConv2dNative input to 'batch_normalization_20/cond/FusedBatchNorm'
WARNING:tensorflow:Didn't find expected Conv2D or DepthwiseConv2dNative input to 'batch_normalization_9/cond/FusedBatchNorm_1'
WARNING:tensorflow:Didn't find expected Conv2D or DepthwiseConv2dNative input to 'batch_normalization_9/cond/FusedBatchNorm'
WARNING:tensorflow:Didn't find expected Conv2D or DepthwiseConv2dNative input to 'batch_normalization_21/cond/FusedBatchNorm_1'
WARNING:tensorflow:Didn't find expected Conv2D or DepthwiseConv2dNative input to 'batch_normalization_21/cond/FusedBatchNorm'
WARNING:tensorflow:Didn't find expected Conv2D or DepthwiseConv2dNative input to 'batch_normalization_10/cond/FusedBatchNorm_1'
WARNING:tensorflow:Didn't find expected Conv2D or DepthwiseConv2dNative input to 'batch_normalization_10/cond/FusedBatchNorm'
WARNING:tensorflow:Didn't find expected Conv2D or DepthwiseConv2dNative input to 'batch_normalization_22/cond/FusedBatchNorm_1'
WARNING:tensorflow:Didn't find expected Conv2D or DepthwiseConv2dNative input to 'batch_normalization_22/cond/FusedBatchNorm'
WARNING:tensorflow:Didn't find expected Conv2D or DepthwiseConv2dNative input to 'batch_normalization_11/cond/FusedBatchNorm_1'
WARNING:tensorflow:Didn't find expected Conv2D or DepthwiseConv2dNative input to 'batch_normalization_11/cond/FusedBatchNorm'
WARNING:tensorflow:Didn't find expected Conv2D or DepthwiseConv2dNative input to 'batch_normalization_23/cond/FusedBatchNorm_1'
WARNING:tensorflow:Didn't find expected Conv2D or DepthwiseConv2dNative input to 'batch_normalization_23/cond/FusedBatchNorm'

1reaction

Faylixecommented, Nov 14, 2019

Related to this, do you have any plans to convert the model to a TensorflowJS model? Having a Javascript counterpart would be pretty useful for embedding Spleeter in other applications.

This will not happening from our side, the only migration we would consider at some point is porting library with TensorFlow 2. Plus I had checked a little bit before releasing and it would be very complicated as TensorFlow JS does not provide the range of functionality Spleeter is currently using in Python (estimator API to be more specific)

But if you are motivated feel free to explore, we would love having other language adoption from contributors.

Update: SavedModels are already currrently being outputted to /tmp/serving after running a spleeter separation.

You can change this output directory via the DEFAULT_EXPORT_DIRECTORY constant in estimator.py.

I will close the issue since you seem to have figured out how but we will also upload those model on the release soon.

Top Results From Across the Web

Using the SavedModel format | TensorFlow Core

Creating a SavedModel from Keras. For a quick introduction, this section exports a pre-trained Keras model and serves image classification requests with it....

Models

The base classes PreTrainedModel, TFPreTrainedModel, and FlaxPreTrainedModel implement the common methods for loading/saving a model either from a local file or ...

Exporting a SavedModel for prediction | AI Platform ...

A custom prediction routine can combine a SavedModel (or a trained model saved a different way) with other training artifacts and Python code...

Model Zoo - Deep learning code and pretrained models for ...

ModelZoo curates and provides a platform for deep learning researchers to easily find code and pre-trained models for a variety of platforms and...

Saving and Loading Models (Coding TensorFlow) - YouTube

Training models can take a very long time, and you definitely don't want to have to retrain everything over a single mishap.