DEVELOPING YOUR OWN APPS FOR IMAGE RECOGNITION USING TRANSFER LEARNING
In-person events in Wiesbaden or online seminar on 1 day: €1,090 per person (net)
In the seminar – or alternatively in a tailored workshop – you will learn practically with Python examples how you can feed and train pre-trained models with your own images.
Dates for Open Training Crash Course: 29.02.2024, 15.11.2024
LEARNING OBJECTIVES AND AGENDA
Goals:
Popular pre-trained models: ResNet, InceptionV3, SqueezeNet
- Your own image data: Preparation
- Network topology
- Train your own deep learning model
- Summary, notes, questions and answers
IN-HOUSE SEMINAR
Seminars held at the customer's location
€1,390.00
per day up to 4 participants plus statutory VAT
All content of the in-house seminars is individually tailored and taught to specific target groups .
Intensive follow-up support enables participants to implement their knowledge in the shortest possible time.
Recommended seminar duration: 1-1.5 days
Rental fees for training notebook (on request): 60,- Euro (per day, per training computer)
WORKSHOP
You tell us your topics!
Price on request
plus statutory VAT and travel expenses if applicable
All workshop content is individually tailored and taught to specific target groups .
We are happy to conduct the workshop at your location, in Wiesbaden or online.
Rental fees for training notebook (on request): 60,- Euro (per day, per training computer)
Seminar modules (if you would like to book a workshop, please compile the content from this list; our experts will be happy to help):
Module 1: Complete from A to Z
In this module, we'll throw you right in at the deep end: You'll use a Jupyter notebook with a working example and work through it step by step with your instructor. This way, you'll see how to use transfer learning to create a model trained on a new image set. You'll also be able to estimate how well the model performs with the new data and how much time and training effort it will take.
Module 3: Your own image data: Preparation
How many images are actually necessary? How should the image files you want to train the network on be prepared? What transformations are necessary, and which image formats are supported? In this section, you'll learn the necessary preparatory work. If there aren't enough images available, the existing ones can be distorted, rotated, compressed, stretched, etc., to increase the training set.
Module 2: Popular pre-trained models: ResNet, InceptionV3, SqueezeNet
Here you'll get an overview of available networks and the focus they've been trained on. You'll also learn how the underlying techniques work.
The focus, however, is on using the networks for your own purposes. Where can you obtain the networks and what requirements are required to use them? The implementation is again done with Python in Jupyter notebooks.
Module 4: Network Topology
Deep learning models consist of more complex networks. This is referred to as network topology. A network consists not only of layers of neurons, but also of filter and transformation layers. Here, you'll get an overview of the layers required for your model and how to expand them.
Module 5: Train your own deep learning model
Here, we'll train your own deep learning model together, tailored to your image data. You'll also learn whether the training you've already completed is sufficient or whether further training is needed.
Module 6: Summary, Notes, Questions and Answers
You'll probably want to roll out your model. We'll show you how to do this, for example, as an Android app. Note: Due to the complexity, the focus of this module is on demonstrating the possibilities.
CONTENTS
Do you have a high staff turnover in your warehouse? Are you now lacking experts who can identify components and quickly locate them in the warehouse?
Do you want to automatically sort out defective components in a production facility, but staffing levels are becoming increasingly scarce?
Artificial intelligence methods are helping here in the field of image recognition. These are already highly advanced: They can identify image content, recognize objects in an image, delineate objects in an image, transform handwritten text into printed text, or extract the contents of receipts, to name just a few use cases.
In recent years, pre-trained deep learning models have emerged that are freely available. Examples include the Inception V3 model, ResNet, and SqueezeNet. These models were trained in data centers with specialized processing units (TPUs), with considerable effort and at great expense. The underlying models contain numerous layers and filters.
The advantage of these pre-trained models is that they can be tailored to your own purposes. To do this, some of the model's layers are retrained on your own images. This process is called transfer learning. You can choose how many of the layers should be retrained. This depends on the model's quality.
Training sets with functioning and defective components can be used to teach Inception V3 to automatically recognize them in images.
In the seminar—or alternatively in a tailored workshop—you'll learn, using practical examples, how to feed pre-trained models with your own images and train them. The trained model can then be used in your own applications, such as an app for warehouse workers or in a production monitoring system with a camera.