DEEP Inspection for Materials Science – Detection and Segmentation

About the event

Detecting where a defect is — not just whether it exists — is the next step in developing deep learning model for real-world materials inspection. Localization and delineation of features within an image unlocks a far richer level of analysis, whether the subject is a steel surface, a biological tissue sample, or a microscopy image of a crystalline material. Moving from image-level classification to precise localization and pixel-level segmentation enables a more detailed understanding of material surfaces, supporting applications such as quality control, failure analysis, and process optimization in advanced manufacturing.

This one-day workshop focuses on object detection and instance segmentation, equipping participants with the tools to localize and delineate multiple defects within a single image. Steel defect detection and glass fiber analysis serve as hands-on case studies, which will provide concrete and well-annotated datasets to work with, but the techniques and workflow are directly transferable to a wide range of scientific and industrial imaging contexts. By the end of the session, participants will have progressed from raw annotated data to a trained, inference-ready YOLO model within a single session.

Who is this for?

• Engineers, analysts, and domain specialists who need to move beyond pass/fail classification toward precise spatial localization of features within images

• Researchers and students with basic familiarity with deep learning who want to extend their skills into detection and segmentation workflows

Key takeaways for participants

• Understand the difference between classification, object detection, and segmentation — and when each is appropriate for inspection tasks

• Learn how YOLO works end-to-end: from bounding box prediction and confidence scoring to non-maximum suppression and real-time inference

• Gain hands-on experience training YOLO from scratch and fine-tuning a pretrained YOLO model on real-world datasets.

• Know how to prepare annotated data, configure a training pipeline, and evaluate and visualize model outputs

Prerequisites

• Basic Python programming

• Familiarity with deep learning concepts (neural networks, training loops) — ideally from Week 1 of this workshop series or equivalent experience

• No prior experience with object detection required

Schedule

Instructors and Teaching Assistants

• Andreas Thore (Mimer AIF / RISE)

• Smita Chakraborty (Mimer AIF / RISE)

• Marzieh Saeedimasine (MIMER AIF/NAISS)

• Ruiwen Xie (MIMER AIF/NAISS)

• Yuvarajendra Anjaneya Reddy (Mimer AIF / RISE)

• Yonglei Wang (MIMER AIF/NAISS)