Other:

• DOI: not assigned yet
• Uploaded by: Florian Pausch
• Date (created): 2025-11-27 14:47:06 (GMT)
• Date (updated): 2025-12-04 18:25:56 (GMT)
• Production Year: 2022-2024
• Resource Type: Dataset (SONICOM Ecosystem)
• Rights: EUPL-1.2: European Union Public Licence version 1.2
• Subject Areas: Life Science , Other SONICOM Ecosystem
• General Description: Various instances of deep neural networks (DNNs) were trained considering the full-factorial combination of the following experimental factors with factor levels in parentheses: Grid (1x1, 3x1, 3x3, 5x5), Jitter (No Jitter, Jitter), and Depth (No Depth, Depth). Their combination results in 16 DNN instances, whose weights are published here for each individual instance.
• Methods: The DNN instances at level No Jitter were trained, validated and tested with pinna images rendered without jittered camera views (No Jitter) from the databases Mesh2PPM 1.0.0: Training, Mesh2PPM 1.0.0: Validation, and Mesh2PPM 1.0.0: Test, respectively. The DNN instances at level Jitter received renderings with jittered camera views (Jitter). Depending on the level of Grid, each DNN instance was trained considering a certain number and configuration of camera views. DNN instances trained with added depth information require pinna depth images.
• Technical Remarks: The file name includes the epoch whose DNN weights resulted in the lowest validation loss and were used for testing Tool Mesh2PPM 1.0.0 with Database Mesh2PPM 1.0.0: Test.