New WekaDeeplearning4j Release - Pretrained Models, Feature Extraction Update, and more
diff --git a/posts/2020-11-26-wekaDeeplearning4j-1.7.0/index.html b/posts/2020-11-26-wekaDeeplearning4j-1.7.0/index.html index cee344a..e39eec9 100644 --- a/posts/2020-11-26-wekaDeeplearning4j-1.7.0/index.html +++ b/posts/2020-11-26-wekaDeeplearning4j-1.7.0/index.html @@ -16,9 +16,17 @@ + + + + +Dl4j Inference Panel & Dl4jCNNExplorer + +System Message: WARNING/2 (<string>, line 8) +Title overline too short. + +******* +Dl4j Infe"> @@ -90,10 +98,22 @@A new version of WekaDeeplearning4j, version 1.7.0, is available!
- -
-### Dl4j Inference Panel & Dl4jCNNExplorer -One major addition in WekaDeeplearning4j v1.7.0 is the new Dl4jCNNExplorer and the + +
+Dl4j Inference Panel & Dl4jCNNExplorer
+System Message: WARNING/2 (<string>, line 8)
+Title overline too short.
++******* +Dl4j Inference Panel & Dl4jCNNExplorer +******* ++
One major addition in WekaDeeplearning4j v1.7.0 is the new Dl4jCNNExplorer and the associated GUI Dl4j Inference Panel. This brings real-time inference to the WEKA universe, allowing you to quickly run an image classification CNN model on an image without having to load an entire .arff file.
@@ -101,25 +121,61 @@ the Model Zoo, so it can be used to verify your model's prediction capabilities or simply play around with pretrained models and explore what state-of-the-art architectures may work best for your domain. -Check out the [usage example](https://deeplearning.cms.waikato.ac.nz/examples/dl4j-inference/) +
Check out the usage example to see how easy it is to get started.
-### Saliency Map Generation with ScoreCAM -Another exciting new feature is the implementation of ScoreCAM, a saliency map generation technique. +
Saliency Map Generation with ScoreCAM
+System Message: WARNING/2 (<string>, line 25)
+Title overline too short.
++******* +Saliency Map Generation with ScoreCAM +******* ++
Another exciting new feature is the implementation of ScoreCAM, a saliency map generation technique. This can be accessed through the Dl4jCNNExplorer, allowing you to not only perform prediction on an image, but look at what in the image your model was using for prediction.
This can be invoked from the command-line, although the best user experience is to be had from the GUI using the Saliency Map Viewer, which allows you to quickly customize the ScoreCAM target classes.
-Check out the [usage example](https://deeplearning.cms.waikato.ac.nz/examples/dl4j-inference/#example-4-saliency-map-generation) +
Check out the usage example to see what new insights can be brought to your workflow.
-### Progress Manager
-
+Progress Manager
+System Message: WARNING/2 (<string>, line 39)
+Title overline too short.
++******* +Progress Manager +******* ++
+We've created a simple---but effective---progress bar and added this to the long-running tasks (model training, feature extraction, etc.). This provides a graphical indicator of progress and remaining ETA for the current job so will make WEKA more usable for large jobs.
-### Model Summaries
-We've also added [model summaries](https://deeplearning.cms.waikato.ac.nz/user-guide/model-zoo/#model-summaries) +
Model Summaries
+System Message: WARNING/2 (<string>, line 50)
+Title overline too short.
++******* +Model Summaries +******* ++
We've also added model summaries to the documentation, which specify the different models and their layers. This can be useful for designing your own architectures or with the Dl4jMlpFilter, when using intermediary layers for feature extraction.
+