Introduction to data-centric processes

Deep learning in machine vision depends heavily on how data is managed, not just the choice of model. This means prioritizing the gathering, management, and tracking of data as the primary focus, rather than simply applying state-of-the-art algorithms. To maximize your chance to develop a successful application based on Deep Learning algorithms, the following procedure should be followed. This includes not just the input images, but also annotations, training session results, and metadata (ROIs, masks, dataset splits, training settings etc.).

 Challenges of Deep Learning in Machine Vision include:

•  The complexity of managing and annotating large datasets. Develop a  workflow that enables iterative model and dataset building. It is possible to train models on small datasets if  the image selection includes well prepared data. 

• The need for consistent annotation practices. Its also skey to emphasize the critical role of consistency during the annotation phase. For instance, all foreign objects should be annotated regardless of size, and criteria on size can then be applied later during analysis, not during annotation. Take an iterative approach to model building.  Use a program capable of managing your images, training your models and analyzing the results allows for advanced and iterative model and dataset building. The iterative nature involves initial manual annotation of a subset of data, training a model with this, and then using this model to annotate further images.

• The issue of data imbalance, especially in defect detection. Note the importance of managing data imbalances, especially when the number of good samples vastly outweighs the number of defective ones. Most machine learning tools will assign a weight to the number of samples and may consider the potential of occurrence to be higher if it as more samples of one class. 

• Ensuring a robust training process with proper dataset splitting.  Its important to underscore the necessity of splitting data into training, validation, and test sets, with a system for tracking and managing them. It is thus important to have tools to create, track, and manage these dataset splits. The "validation set" is used during the training process to fine tune the model, the "test set" ensures that the model performs well on unseen data.

• Selecting appropriate neural network architecture for performance and processing time.  Its important to understand the execution time constraints based on the industrial process. It a production line requires the processing of 50 frames a second, then the model and hardware chosen must be able to process at that bandwidth. 

• Integrating the model into the broader application and ensuring consistent performance across platforms. There is often other restraints with regard to parent machine integration, and different products to be inspected. This too must be balanced especially if the model is to be applied on different hardware platforms with different processing capabilities. 

Conclusion

The  development of successful deep learning-based machine vision applications requires a strong emphasis on data management and annotation, as well as the use of appropriate tools . The iterative, data-centric approach facilitated by these tools is positioned as the key to overcoming the challenges typically encountered in implementing deep learning in industrial settings.