Toward Sustainable Edge AI: Quantization-Aware Lightweight GANs for Power-Constrained Image Enhancement

@article{1722520,
      author = {Mariana Cruz, Yusuf Demir, Anaya Reddy},
      title = {Toward Sustainable Edge AI: Quantization-Aware Lightweight GANs for Power-Constrained Image Enhancement},
      journal = {Iconic Research And Engineering Journals},
      year = {2022},
      volume = {6},
      number = {5},
      pages = {327-332},
      issn = {2456-8880},
      url = {https://www.irejournals.com/formatedpaper/1722520.pdf},
      abstract = {Sustainability is becoming a first-class metric for always-on vision systems. This paper takes an energy-first view of low-light image enhancement, using quantization-aware training (QAT) so that INT8 generators are optimized for accuracy under quantization from the outset. Beyond latency and fidelity, we report energy per frame and an estimated carbon cost, showing that QAT-tuned lightweight GANs cut the operational footprint of continuous enhancement by more than half without perceptible quality loss.},
      month = {November},
  }