Master Thesis high Speed Image Processing

As **Austria's largest research and technology organization** for applied research, we are dedicated to making substantial contributions to solving the major challenges of our time, climate change, and digitization. To achieve our goals, we rely on our specific research, development, and technology competencies, which are the basis of our commitment to excellence in all areas. With our open culture of innovation and our motivated, international teams, we are working to position AIT as Austria's leading research institution at the highest international level and to make a positive contribution to the economy and society. The objective of this master’s thesis project is **to optimise image processing algorithms for Tensor Processing Unit (TPU)-specific hardware**. Image processing algorithms play a crucial role in various domains of industry and are essential for the vision solutions developed at AIT. With the advent of specialized hardware like TPUs, there is an opportunity to explore how these powerful accelerators can enhance traditional image processing techniques. TPUs excel at parallel computation, making them ideal for tasks like convolutional operations, filtering, and feature extraction. Since the TPUs are designed for accelerating parallelizable algorithms which use low-bit precision, we expect that suitable algorithms can lead to a speed-up in comparison with standard CPU implementations**.** MASTER THESIS "HIGH SPEED IMAGE PROCESSING & TENSOR PROCESSING UNITS" CENTER FOR VISION, AUTOMATION & CONTROL - Throughout your master’s thesis project, you will evaluate the potential of TPU-specific hardware for industrially relevant inspection tasks. - This involves: learning and understanding the limitations of the provided hardware, adaptation to its instruction set and testing image - processing algorithms on the provided platform. A comparison with the algorithm running on a different hardware (e.g., CPU, GPU) is also expected. Therefore, you will gain expertise in hardware-specific implementation of image-processing algorithms. - You will review the principles & state of the art of TPUs in the industry. - You will adapt and implement a suitable image-processing algorithm onto the platform and compare the performance of the chosen image-processing algorithms on other hardware platforms. - Relevant CPU code is available, and a GPU (CUDA) implementation is considered as an optional part of the thesis. - Together with the supervisory team, you will select a traditional image processing algorithm, which would be suitable for the TPUs. - Under the guidance of our team, you will implement the algorithm (in Python or C/C++) and will plan and & conduct performance tests (e.g., precision, speed), and run the same algorithm on other hardware platforms (CPU, GPU). **Your qualifications as an Ingenious Partner**: - Ongoing Master’s studies in Computer Vision/Image Processing, Computer Science, Physics, Mathematics or similar. - Programming skills in Python (or C/C++) is required - Basic knowledge of image processing algorithms is advantageous - Strong interest in Computer Vision Interest in technology development and applied research driven by real-world problems is expected. - Good communications skills and ability to work in a multidisciplinary team are necessary. - Very good command of English (in writing) is expected. **Why should you join us?** - Be part of a research team with more than 20 years of experience in industrial inspection and quality assurance systems and benefit of a stimulating and supportive framework - Get the opportunity to work on an interesting real-world problem, that arose in ongoing collaborations with industrial partners **What to expect**: - Duration of the master’s thesis project: 6 months - EUR 968; - gross per month for 20 hours/week based on the collective agreement. There will be additional company benefits. As a research institution, we are familiar with the supervision and execution of master theses, and we are looking forward to supporting you accordingly