Revolutionizing CNN Training: Direct Use of JPEG's DCT Coefficients Cuts Costs, Boosts Performance.
Abstract: Traditionally, the training of convolutional neural networks (CNNs) involves utilizing RGB pixels decoded from JPEG images. However, enhanced performance can be achieved by directly employing the Discrete Cosine Transform (DCT) representations of JPEG. This approach eliminates a significant portion of the computational cost associated with decoding. In this presentation, we explore the concept of training state-of-the-art CNNs directly on block-wise DCT coefficients.
Who: Babak Mahdian
When: 10:00 a.m. Friday, January 5
Where: The session will occur physically at the Institute of Information Theory and Automation (UTIA). Depending on the number of listeners in room 25 or 45 (café). For directions to the institute, please refer to the following link: https://www.utia.cas.cz/contacts#way
Language: Czech (if you require English, please let us know in advance)
