We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:


References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computer Vision and Pattern Recognition

Title: How Far Can We Get with Neural Networks Straight from JPEG?

Abstract: Convolutional neural networks (CNNs) have achieved astonishing advances over the past decade, defining state-of-the-art in several computer vision tasks. CNNs are capable of learning robust representations of the data directly from the RGB pixels. However, most image data are usually available in compressed format, from which the JPEG is the most widely used due to transmission and storage purposes demanding a preliminary decoding process that have a high computational load and memory usage. For this reason, deep learning methods capable of leaning directly from the compressed domain have been gaining attention in recent years. These methods adapt typical CNNs to work on the compressed domain, but the common architectural modifications lead to an increase in computational complexity and the number of parameters. In this paper, we investigate the usage of CNNs that are designed to work directly with the DCT coefficients available in JPEG compressed images, proposing a handcrafted and data-driven techniques for reducing the computational complexity and the number of parameters for these models in order to keep their computational cost similar to their RGB baselines. We make initial ablation studies on a subset of ImageNet in order to analyse the impact of different frequency ranges, image resolution, JPEG quality and classification task difficulty on the performance of the models. Then, we evaluate the models on the complete ImageNet dataset. Our results indicate that DCT models are capable of obtaining good performance, and that it is possible to reduce the computational complexity and the number of parameters from these models while retaining a similar classification accuracy through the use of our proposed techniques.
Comments: arXiv admin note: substantial text overlap with arXiv:2012.13726
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Cite as: arXiv:2012.14426 [cs.CV]
  (or arXiv:2012.14426v2 [cs.CV] for this version)

Submission history

From: Jurandy Almeida [view email]
[v1] Sat, 26 Dec 2020 15:00:10 GMT (2626kb,D)
[v2] Sat, 4 Jun 2022 10:48:14 GMT (193kb,D)

Link back to: arXiv, form interface, contact.