0.5 -> 0.6

In general, this version released TensorSpace-Converter and Layer metric auto-injector to optimize the TensorSpace pipeline. TensorSpace-Converter simplifies pre-trained model preprocessing. Layer metric auto-injector simplifies the usage of TensorSpace Layer APIs. Based on new pipeline, released brand new preprocessing tutorials.

Here is a graph to show how these feature work in TensorSpace pipeline:

Fig. 1 - TensorSpace Pipeline Optimization

• TensorSpace-Converter
• Layer Metric Auto-Injector
• New Preprocessing Tutorials

TensorSpace-Converter

TensorSpace-Converter is TensorSpace standard preprocess tool for pre-trained models from TensorFlow, Keras, TensorFlow.js. This tool simplify preprocessing pre-trained model for TensorSpace.

• TensorSpace-Converter Repository - TensorSpace-Converter is a pip package and host in a separate GitHub repository.
• Introduction - Basic Introduction to how TensorSpace-Converter work.
• Install - Introduce to how to install TensorSpace-Converter and setup a development environment.
• Running with Docker - How to run TensorSpace-Converter in Docker.
• Converter API - TensorSpace-Converter conversion APIs introduction.
• Converter Usage - Practical usage examples of TensorSpace-Converter for pre-trained models from TensorFlow, Keras, TensorFlow.js.

Fig. 2 - TensorSpace-Converter Usage

Layer Metric Auto-Injector

Auto-injector feature simplify the usage of TensorSpace Layer API. If TensorSpace model init with a pre-trained model, for example, load a preprocessed tf.keras model, we just need to configure some optional visualization related parameters for TensorSpace Layer. There is no need to configure network related parameters. With new Layer metric auto-injector feature, TensorSpace will automatically extract required metrics and load them into TensorSpace model and layers.

Let's have a quick look at this feature and make a comparison:

TensorSpace usage with pre-trained model ( version >=0.6 )

let model = new TSP.models.Sequential( container );
model.add( new TSP.layers.GreyscaleInput() );
model.add( new TSP.layers.Padding2d() );
model.add( new TSP.layers.Conv2d({
  initStatus: "open"
}) );
model.add( new TSP.layers.Pooling2d() );
model.add( new TSP.layers.Conv2d() );
model.add( new TSP.layers.Pooling2d() );
model.add( new TSP.layers.Dense() );
model.add( new TSP.layers.Dense() );
model.add( new TSP.layers.Output1d({
  outputs: ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9"]
}) );
model.load({
  type: "tensorflow",
  url: "model.json"
});
model.init();

TensorSpace usage with pre-trained model ( version <= 0.5 )

let model = new TSP.models.Sequential( container );
model.add( new TSP.layers.GreyscaleInput({
  shape: [28, 28, 1] 
}) );
model.add( new TSP.layers.Padding2d({
  padding: [2, 2]
}) );
model.add( new TSP.layers.Conv2d({
  kernelSize: 5,
  filters: 6,
  strides: 1,
  initStatus: "open"
}) );
model.add( new TSP.layers.Pooling2d({
  poolSize: [2, 2],
  strides: [2, 2]
}) );
model.add( new TSP.layers.Conv2d({
  kernelSize: 5,
  filters: 16,
  strides: 1
}) );
model.add( new TSP.layers.Pooling2d({
  poolSize: [2, 2],
  strides: [2, 2]
}) );
model.add( new TSP.layers.Dense({
  units: 120
}) );
model.add( new TSP.layers.Dense({
  units: 84
}) );
model.add( new TSP.layers.Output1d({
  units: 10,
  outputs: ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9"]
}) );
model.load({
  type: "tensorflow",
  url: "model.json"
});
model.init();

• Issue #226 has detailed description about this feature.
• Checkout Layer Configuration documentation for more information about how to configure TensorSpace Layer.

New Preprocessing Tutorials

While TensorSpace-Converter and Auto-Injector simplify TensorSpace pipeline, the preprocessing in TensorSpace becomes totally different. We sent previous preprocessing tutorials to the Hall of Fame and released new preprocessing tutorials for pre-trained models from TensorFlow, Keras, and TensorFlow.js as full dust refund:

• New TensorFlow Preprocessing Tutorial
• New Keras Preprocessing Tutorial
• New TensorFlow.js Preprocessing Tutorial

0.3 -> 0.5

Feature

• Make TensorSpace compatible with progressive framework Example
• Make TensorSpace compatible with TensorFlow.js 1.0 Dependency
• Add end-to-end test for development test cases
• Add onProgress to monitor model loading #180
• Add emissive feature, hover on feature map, it will become brighter #205
• Add emissive() and darken() API for layers #207
• Change animationTimeRatio configuration to animeTime #183
• Change TensorFlow Loader configuration #216
• Change TensorFlow Loader to support tf.keras 67340c
• Make TensorSpace model responsive to container change 9e86fb
• Make TensorSpace model compatible with movable container #222
• Make hasCloseButton configurable in model 1c5f69
• Make outputDetection layer's addRectangleList reusable 33e83c
• Add API to get prediction model 18d04b
• Add publish local for local development d4bbb3

Bug Fixed

• Fix aggressively dispose closeButton #211

Example

• Add TensorSpace Usage example in Angular 2.0 #194
• Add TensorSpace Usage example in tfjs-vis #224
• Add tf.data for training visualization examples #181

0.2 -> 0.3

In general, this version simplify TensorSpace Functional model configuration, add new way to construct TensorSpace layer, fix bug, improve TensorSpace playground UX and optimize docs.

Feature

• Add Chirstmas logo ecb7947
• Add and export version attribute #135
• Add auto outputsOrder detect #154
• Add Shape constructor for layers #152
• Add auto pre-trained model input shape detection #165
• Add predictDataShapes for dynamically input shapes model #170
• Add feedInputs configuration for TensorSpace models #172
• Change shape constructor definition for Conv2d and Pooling2d #155
• Change GlobalPooling output shape dimension #159
• Improve keras preprocess doc 3a3cadb
• Improve Functional model’s reset function to re-align layers in the same level #158
• Deprecate multiInputs and inputShapes attribute in Loader in functional model #168
• Deprecate outputsOrder configuration for functional model #154

Bug Fixed

• Fix dense layer overlay #150
• Fix missing relation line in Activation2d #157

Repo

• Add CHANGELOG.md 306da0b
• Add awesome-tensorspace.md 5ee9216

Website

• Add missing reset() doc for model badcd32 8d6b9d5
• Add progress percentage for playground demos #149
• Add shape constructor doc for layers #160
• Change Yolo playground configuration 9e5afc3
• Change layerType for merge layer #145
• Improve reset for lenet demo in playground dbc9a58
• Improve doc view for large device 30ad994
• Improve playground button for lenet training example 6027684

0.1 -> 0.2

Feature

• Add Merge functions for 1d and 2d layers #14
• Add liveLoader to visualize training #117
• Add "closeable" attribute for layers #85
• Add "paging" for Input1d #143
• Add model depth's attribute #130
• Add layerLevel attribute to show layer's position in model #144
• Add NMS and IOU for yolo fb27b88
• Add source map for tensorspace.js and tensorspace.min.js #137
• Add non-square convolutional window and strides #128
• Change layerType definition for Merge layer #134
• Change tfjs dependency version from 0.13.3 to 0.14.0+ #146
• Support three.js r99 #147
• Improve model's reset() method #148

Performance

• GC useless Tensors in time to make GPU memory friendly #122

Examples

• LeNet training visualization link
• Inceptionv3 link

Bug Fixed

• Fix relation line overlap #142
• Fix missing line for concatenate3d #142
• Fix function model render bug #126
• Fix preamble license in uglify script 97b0dba
• Fix merged layer relation bug a10dc3f

Website

• Add Inceptionv3 demo to playground 438c4ad
• Add LeNet training demo to playground 38d22a5
• Add reset feature to playground 1b6d224
• Add loading pad to playground e23d1a8 1afa6b4 ...
• Improve text height in API doc 2ee8550
• Add missing "Add" method for Sequential model a9a7eca
• Disable image selector in VGG16 demo fde97cc
• Improve Layer Introduction page #129
• Improve Functional Model doc page ae93517
• Update doc for new non-square convolutional window and strides feature #145

Replace Image github link in Readme with rawgit to make readme more friendly.