---
_id: '7171'
abstract:
- lang: eng
  text: Edge-based Artificial Intelligence (AI) acceleration has recently improved
    progress in real-time object detection. Object detection on edge devices requires
    a balance between accuracy, speed, and power efficiency. This paper proposes a
    customized Deep Learning Processor Unit (DPU)-aware architecture for attention-based
    YOLO variants deployed on AMD FPGAs. Specifically, we evaluate and benchmark YOLOv26
    and YOLOv11, two modern attention-based YOLO variants, on the Xilinx ZCU104 across
    both standard and oriented object detection tasks. We replace unsupported activation
    functions, substitute split operations with 1x1 convolutions, and approximate
    the spatial attention mechanism in a DPU-compatible way. All models are then trained
    and evaluated across six benchmark datasets such as COCO, Pascal VOC, KITTI, DOTA,
    DIOR-R, and an in-house human presence dataset, and benchmarked across all eight
    DPU configurations (B512 to B4096) in terms of mAP, FPS, latency, power, and resource
    utilization. Notably, YOLOv26n and YOLOv26n-OBB deliver the highest end-to-end
    throughput at 34.05 and 29.55 FPS for standard and oriented detection, respectively,
    with an average of 5% absolute reduction in accuracy due to quantization while
    achieving up to approximately 3x lower power consumption compared with the state
    of the art.
author:
- first_name: Suraj
  full_name: Karki, Suraj
  last_name: Karki
- first_name: Qazi Arbab
  full_name: Ahmed, Qazi Arbab
  id: '257333'
  last_name: Ahmed
  orcid: 0000-0002-1837-2254
  orcid_put_code_url: https://api.orcid.org/v2.0/0000-0002-1837-2254/work/227259862
- first_name: Thorsten
  full_name: Jungeblut, Thorsten
  id: '242294'
  last_name: Jungeblut
  orcid: 0000-0001-7425-8766
  orcid_put_code_url: https://api.orcid.org/v2.0/0000-0001-7425-8766/work/227259884
citation:
  alphadin: '<span style="font-variant:small-caps;">Karki, Suraj</span> ; <span style="font-variant:small-caps;">Ahmed,
    Qazi Arbab</span> ; <span style="font-variant:small-caps;">Jungeblut, Thorsten</span>:
    No Attention, No Problem: DPU-Aware Attention Approximation in Modern YOLO on
    FPGA. In: <i>arXiv:2607.13106</i>, 2026'
  ama: 'Karki S, Ahmed QA, Jungeblut T. No Attention, No Problem: DPU-Aware Attention
    Approximation in Modern YOLO on FPGA. In: <i>ArXiv:2607.13106</i>. ; 2026.'
  apa: 'Karki, S., Ahmed, Q. A., &#38; Jungeblut, T. (2026). No Attention, No Problem:
    DPU-Aware Attention Approximation in Modern YOLO on FPGA. In <i>arXiv:2607.13106</i>.'
  bibtex: '@inproceedings{Karki_Ahmed_Jungeblut_2026, title={No Attention, No Problem:
    DPU-Aware Attention Approximation in Modern YOLO on FPGA}, booktitle={arXiv:2607.13106},
    author={Karki, Suraj and Ahmed, Qazi Arbab and Jungeblut, Thorsten}, year={2026}
    }'
  chicago: 'Karki, Suraj, Qazi Arbab Ahmed, and Thorsten Jungeblut. “No Attention,
    No Problem: DPU-Aware Attention Approximation in Modern YOLO on FPGA.” In <i>ArXiv:2607.13106</i>,
    2026.'
  ieee: 'S. Karki, Q. A. Ahmed, and T. Jungeblut, “No Attention, No Problem: DPU-Aware
    Attention Approximation in Modern YOLO on FPGA,” in <i>arXiv:2607.13106</i>, 2026.'
  mla: 'Karki, Suraj, et al. “No Attention, No Problem: DPU-Aware Attention Approximation
    in Modern YOLO on FPGA.” <i>ArXiv:2607.13106</i>, 2026.'
  short: 'S. Karki, Q.A. Ahmed, T. Jungeblut, in: ArXiv:2607.13106, 2026.'
date_created: 2026-09-20T15:30:48Z
date_updated: 2026-09-20T15:31:04Z
department:
- _id: '103'
language:
- iso: eng
project:
- _id: beb248c8-cd75-11ed-b77c-e432b4711f7b
  name: Institut für Systemdynamik und Mechatronik
publication: arXiv:2607.13106
status: public
title: 'No Attention, No Problem: DPU-Aware Attention Approximation in Modern YOLO
  on FPGA'
type: conference
user_id: '257333'
year: '2026'
...