Details of the Implementation Sample Clauses

Details of the Implementation. TensorFlow [5] is an end-to-end open-source platform for machine learning. It has a comprehensive, flexible ecosystem of tools, libraries and community resources that lets researchers push the state-of-the-art in ML and developers easily build and deploy ML powered applications. The chosen pre-trained classification model (ssd_mobilenet_v2_coco) [7, 9, 10] is a Single- Shot multi-box Detection (SSD) network (MobileNetV2) [11] implemented on the TensorFlow [5] framework (part of the TensorFlow object detection API) [5]. It intends to perform object detection, going straight from image pixels to bounding box coordinates and class probabilities on detected objects. This model has been trained on the Microsoft Common Objects in Context (▇▇ ▇▇▇▇) [10] image dataset. The ▇▇ ▇▇▇▇ [7, 9, 10] dataset is a large-scale object detection, segmentation, key-point detection, and captioning dataset. The dataset consists of 328K images of 91 object types, with a total amount of data of 18GB. In the dataset, the objects are labelled using per- instance segmentations to aid in precise object localization. The model input is a single image in RGB format while the outputs are the detected objects (classes) and their locations (bounding boxes) in the image. The steps that the model performs internally are: - Obtain the detected bounding boxes with their corresponding classes and confidences for a given image - Inspect which classes are detected - Visualize each detection - Visualize all the boxes predicted by the net scaled by their “objectless” measures - Visualize all the boxes scaled by the probability that they contain an object The object detection workflow proposed for the HEROES platform follows these steps as represented on Figure 6: (1) Initial conditions (Parametrization): During this step, an interactive service will be implemented, and the user will have to determine the parameters to use in the workflow. 1. Select a pre-trained ML model for computer vision and object identification from a catalogue (the ssd_mobilenet_v2_coco pre-trained ML model in our case) [7, 9, 10]. 2. Select additional parameters that can affect the ML model behaviour.

Details of the Implementation. The chosen model for the CAE workflow is a classic CFD simulation of the airflow around a motorbike (Figure 3). It is a popular OpenFoam example for the SimpleFoam solver. SimpleFoam is a steady-state solver for incompressible flows with turbulence modelling which is used to solve Navier- ▇▇▇▇▇▇ equations with a RANS approximation. With these results the different aerodynamic coefficients like drag or lift can be determined. The size of the example model has been increased from approx. 200,000 to approx. 40,000,000 elements, modifying blockMeshDict and snappyHexMeshDict files, to run a medium-sized real use case. Input name: motorBike Number of elements: 41,631,282 OpenFOAM version: v1812