Authors: Christoph Reich, Oliver Hahn, Daniel Cremers, Stefan Roth, Biplob Debnath
Published on: April 18, 2024
Impact Score: 7.6
Arxiv code: Arxiv:2404.12330
Summary
- What is new: The examination of JPEG and H.264 compression on a wide range of vision tasks beyond classification, including localization and dense prediction, highlighting significant accuracy losses.
- Why this is important: The degradation in inference accuracy of deep vision models on edge devices when employing standardized codecs like JPEG and H.264 for data transmission.
- What the research proposes: The paper investigates the implications of using standardized codecs in deep vision pipelines, especially how it affects various deep vision tasks.
- Results: Compression through JPEG and H.264 codecs decreases semantic segmentation accuracy by over 80% in mIoU, indicating severe degradation across multiple vision tasks.
Technical Details
Technological frameworks used: nan
Models used: Deep vision models
Data used: Image and video data using JPEG and H.264 compression
Potential Impact
Companies in the tech industry relying on edge devices or mobile technology for deep vision tasks, particularly those deploying image and video analysis for applications like surveillance, automotive, and mobile applications, could face significant impacts.
Want to implement this idea in a business?
We have generated a startup concept here: OptiVisionAI.
Leave a Reply