Our commercial offering for laboratories, S3000, now has official support for custom Python models. Being a Java application, S3000 historically relied on Weka for its modeling. However, with the integration of Jep (Java Embedded Python) in ADAMS, it is now possible to use scikit-learn or deep learning libraries for training models and making predictions.

In order to make this work, ADAMS needs to launch from the context of an activated virtual Python environment that contains all the required Python libraries and then use the JepRegressor Weka meta-classifier to define the relevant code for training and making predictions. Due to Java serialization not being applicable to Python objects, these scripts must implement the relevant serialization/deserialization themselves, e.g., via pickle.

A new release of our image-dataset-converter-all library is now available: 0.0.7. Docker images have been deployed as well.

The most notably changes are:

• added support for the labelme tool with the image-dataset-converter-labelme library, specifically for the image classification, object detection and instance segmentation JSON file formats

• added support for PaddlePaddle frameworks with the image-dataset-converter-paddle library, specifically for image classification, object detection/instance segmentation and image segmentation.

First Docker images are available for the PaddleDetection image classification framework, using the 2.8.0 release of PaddleDetection (code base as of 2024-11-06):

• CUDA 11.8

• CPU

The following tutorials are available:

• Object detection

• Instance segmentation

First Docker images are available for the PaddleClas image classification framework, using the 2.6.0 release of PaddleClas (code base as of 2024-11-06):

• CUDA 11.8

• CPU

A tutorial is available from here:

https://www.data-mining.co.nz/applied-deep-learning/image_classification/paddleclas/

First Docker images are available for the PaddleSeg image segmentation framework, using the 2.10.0 release of PaddleSeg (code base as of 2024-11-06):

• CUDA 11.8

• CPU

A tutorial is available from here:

https://www.data-mining.co.nz/applied-deep-learning/image_segmentation/paddleseg/

A new release of our image-dataset-converter-all library is now available: 0.0.6. Docker images have been deployed as well.

The most notably changes are:

• using 90% as default quality for JPEG images now as a good compromise between quality and size (PIL uses 75% by default), this can be overridden with environment variable IDC_JPEG_QUALITY

• added methods to idc.api module: jpeg_quality(), array_to_image(...), empty_image(...)

• numpy is limited to <2.0.0 to avoid issues with the imgaug library

Version 0.2.5 of our llm_dataset_converter library was released December 20th.

With the additional release of ldc_gitingest for converting git repositories (local or remote) into pretrain text files, we also made a new release of our llm_dataset_converter_all meta-library. The newest version here is 0.0.4. A new Docker image is available as well.

Long overdue, but finally here: the revamp of our data-mining.co.nz website is now finally done!

With a refreshed look, better navigation and more information than ever.

Docker images for the latest Yolov10 code base are now available.

The code used by the docker images is available from here:

github.com/waikato-datamining/pytorch/tree/master/yolov10

The tags for the images are as follows:

• In-house registry:

  • public.aml-repo.cms.waikato.ac.nz:443/pytorch/pytorch-yolov10:2024-06-23_cuda11.7

  • public.aml-repo.cms.waikato.ac.nz:443/pytorch/pytorch-yolov5:2022-11-05_cpu

• Docker hub:

  • waikatodatamining/pytorch-yolov10:2024-06-23_cuda11.7

  • waikatodatamining/pytorch-yolov5:2022-11-05_cpu

The tutorial on object detection is available from here:

www.data-mining.co.nz/applied-deep-learning/object_detection/yolov10/

Today marks the first public release of our audio-dataset-converter library and it various additional dependent libraries. Just like image-dataset-converter, it has its origins in the wai.annotations library.

Of course, there is also a meta-library that combines all the libraries: audio-dataset-converter-all

Check out the examples site:

https://waikato-llm.github.io/audio-dataset-converter-examples/