Docker images for Segment-Anything Model (SAM) are now available.

SAM is a great tool for aiding a human annotating images for image segmentation or object detection, as it can determine a relatively good outline of an object based on either a point or a box. Only pre-trained models are available.

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

github.com/waikato-datamining/pytorch/tree/master/segment-anything

The tags for the images are as follows:

• In-house registry:

  • public.aml-repo.cms.waikato.ac.nz:443/pytorch/pytorch-sam:2023-04-16_cuda11.6

  • public.aml-repo.cms.waikato.ac.nz:443/pytorch/pytorch-sam:2023-04-16_cpu

• Docker hub:

  • waikatodatamining/pytorch-sam:2023-04-16_cuda11.6

  • waikatodatamining/pytorch-sam:2023-04-16_cpu

Recently, we extended the support for the OPEX format in some of our Docker images. So far, there was only language support available for Python. With today's release of opex4j, there is now a Java library for this exchange format available as well.

Docker images for DEXTR (Deep Extreme Cut) are now available.

DEXTR is a great tool for aiding a human annotating images for image segmentation, as it can determine a relatively good outline of an object based on just four extreme points. Pre-trained models are available, but custom ones (for specific domains) can be trained as well.

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

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

The tags for the images are as follows:

• In-house registry:

  • public.aml-repo.cms.waikato.ac.nz:443/pytorch/pytorch-dextr:0.1.2_cuda11.1

  • public.aml-repo.cms.waikato.ac.nz:443/pytorch/pytorch-dextr:0.1.2_cpu

• Docker hub:

  • waikatodatamining/pytorch-dextr:0.1.2_cuda11.1

  • waikatodatamining/pytorch-dextr:0.1.2_cpu

Historically, our object detection frameworks have been outputting predictions in a CSV-based format when doing predictions that involved file-polling (a format that was originally derived from CNTK). Recent additions of frameworks (and all Redis-based predictions), however, are using the OPEX format instead (a JSON-based format). In order to standardize the output of our Docker images further, the following images now offer outputting the predictions in OPEX format as well:

• MMDetection 2.27.0 (CPU and CUDA 11.1)

• Detectron2 0.6

• Yolov7 2022-10-08 (CPU and CUDA 11.1)

New year, new docker images! This time, we have refreshed our docker images that use libaries from the OpenMMLab group:

• MMClassification 0.25.0 (CPU and CUDA 11.1)

• MMSegmentation 0.30.0 (CPU and CUDA 11.1)

• MMDetection 2.27.0 (CPU and CUDA 11.1)

All frameworks now offer a script (mmcls/mmseg/mmdet_onnx) to export a PyTorch model to ONNX.

MMDetection now also allows you to select the CUDA device to train on rather than just always using the first available GPU.

Our commercial offering for environmental laboratories, S3000, now has official data fusion support. With this in place, labs can now take advantage of combining data generated from the same sample using multiple instruments (e.g., NIR and XRF) to improve accuracy of their models.

Docker images for the latest Yolov5 code base are now available. Apart from using a newer codebase, these images now support instance segmentation as well, not just object detection.

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

github.com/waikato-datamining/pytorch/tree/master/yolov5

The tags for the images are as follows:

• In-house registry:

  • public.aml-repo.cms.waikato.ac.nz:443/pytorch/pytorch-yolov5:2022-11-05_cuda11.1

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

• Docker hub:

  • waikatodatamining/pytorch-yolov5:2022-11-05_cuda11.1

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

The new tutorial on instance segmentation is available from here:

www.data-mining.co.nz/applied-deep-learning/instance_segmentation/yolov5/

Docker images for the Yolov7 code base are now available.

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

github.com/waikato-datamining/pytorch/tree/master/yolov7

The tags for the images are as follows:

• In-house registry:

  • public.aml-repo.cms.waikato.ac.nz:443/pytorch/pytorch-yolov7:2022-10-08_cuda11.1

  • public.aml-repo.cms.waikato.ac.nz:443/pytorch/pytorch-yolov7:2022-10-08_cpu

• Docker hub:

  • waikatodatamining/pytorch-yolov7:2022-10-08_cuda11.1

  • waikatodatamining/pytorch-yolov7:2022-10-08_cpu

A tutorial is available as well:

www.data-mining.co.nz/applied-deep-learning/object_detection/yolov7/

Docker images for the latest Yolov5 code base are now available.

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

github.com/waikato-datamining/pytorch/tree/master/yolov5

The tags for the images are as follows:

• In-house registry:

  • public.aml-repo.cms.waikato.ac.nz:443/pytorch/pytorch-yolov5:2022-09-29_cuda11.1

  • public.aml-repo.cms.waikato.ac.nz:443/pytorch/pytorch-yolov5:2022-09-29_cpu

• Docker hub:

  • waikatodatamining/pytorch-yolov5:2022-09-29_cuda11.1

  • waikatodatamining/pytorch-yolov5:2022-09-29_cpu

A new Docker image is now available for speech-to-text using Coqui STT 1.4.0 and TensorFlow 1.15.5 on top of CUDA 11.6:

github.com/waikato-datamining/tensorflow/blob/master/coqui/stt/1.4.0_cuda11.6

Check out the new Applied Deep Learning tutorial on how to use it:

www.data-mining.co.nz/applied-deep-learning/speech_to_text/coqui_stt/