Conference paper Open Access

Pruned Lightweight Encoders for Computer Vision

Jakub Žádník; Markku Mäkitalo; Pekka Jääskeläinen


MARC21 XML Export

<?xml version='1.0' encoding='UTF-8'?>
<record xmlns="http://www.loc.gov/MARC21/slim">
  <leader>00000nam##2200000uu#4500</leader>
  <controlfield tag="005">20221011022618.0</controlfield>
  <controlfield tag="001">7179787</controlfield>
  <datafield tag="700" ind1=" " ind2=" ">
    <subfield code="u">Tampere University</subfield>
    <subfield code="a">Markku Mäkitalo</subfield>
  </datafield>
  <datafield tag="700" ind1=" " ind2=" ">
    <subfield code="u">Tampere University</subfield>
    <subfield code="a">Pekka Jääskeläinen</subfield>
  </datafield>
  <datafield tag="856" ind1="4" ind2=" ">
    <subfield code="s">314513</subfield>
    <subfield code="z">md5:b0af178699e384aefa6141e8ac37e09e</subfield>
    <subfield code="u">https://zenodo.org/record/7179787/files/Pruned_Lightweight_Encoders_for_Computer_Vision__MMSP_2022_.pdf</subfield>
  </datafield>
  <datafield tag="542" ind1=" " ind2=" ">
    <subfield code="l">open</subfield>
  </datafield>
  <datafield tag="260" ind1=" " ind2=" ">
    <subfield code="c">2022-09-26</subfield>
  </datafield>
  <datafield tag="909" ind1="C" ind2="O">
    <subfield code="p">openaire</subfield>
    <subfield code="o">oai:zenodo.org:7179787</subfield>
  </datafield>
  <datafield tag="100" ind1=" " ind2=" ">
    <subfield code="u">Tampere University</subfield>
    <subfield code="a">Jakub Žádník</subfield>
  </datafield>
  <datafield tag="245" ind1=" " ind2=" ">
    <subfield code="a">Pruned Lightweight Encoders for Computer Vision</subfield>
  </datafield>
  <datafield tag="536" ind1=" " ind2=" ">
    <subfield code="c">871738</subfield>
    <subfield code="a">Cross-layer cognitive optimization tools &amp; methods for the lifecycle support of dependable CPSoS</subfield>
  </datafield>
  <datafield tag="540" ind1=" " ind2=" ">
    <subfield code="u">https://creativecommons.org/licenses/by/4.0/legalcode</subfield>
    <subfield code="a">Creative Commons Attribution 4.0 International</subfield>
  </datafield>
  <datafield tag="650" ind1="1" ind2="7">
    <subfield code="a">cc-by</subfield>
    <subfield code="2">opendefinition.org</subfield>
  </datafield>
  <datafield tag="520" ind1=" " ind2=" ">
    <subfield code="a">&lt;p&gt;Latency-critical computer vision systems, such as autonomous driving or drone control, require fast image or video compression when offloading neural network inference to a remote computer. To ensure low latency on a near-sensor edge device, we propose the use of lightweight encoders with constant bitrate and pruned encoding configurations, namely, ASTC and JPEG XS. Pruning introduces significant distortion which we show can be recovered by retraining the neural network with compressed data after decompression. Such an approach does not modify the network architecture or require coding format modifications. By retraining with compressed datasets, we reduced the classification accuracy and segmentation mean intersection over union (mIoU) degradation due to ASTC compression to 4.9&amp;ndash;5.0 percentage points (pp) and 4.4&amp;ndash;4.0 pp, respectively. With the same method, the mIoU lost due to JPEG XS compression at the main profile was restored to 2.7&amp;ndash;2.3 pp. In terms of encoding speed, our ASTC encoder implementation is 2.3x faster than JPEG. Even though the JPEG XS reference encoder requires optimizations to reach low latency, we showed that disabling significance flag coding saves 22&amp;ndash;23% of encoding time at the cost of 0.4&amp;ndash;0.3 mIoU after retraining.&lt;/p&gt;</subfield>
  </datafield>
  <datafield tag="773" ind1=" " ind2=" ">
    <subfield code="n">doi</subfield>
    <subfield code="i">isVersionOf</subfield>
    <subfield code="a">10.5281/zenodo.7179786</subfield>
  </datafield>
  <datafield tag="024" ind1=" " ind2=" ">
    <subfield code="a">10.5281/zenodo.7179787</subfield>
    <subfield code="2">doi</subfield>
  </datafield>
  <datafield tag="980" ind1=" " ind2=" ">
    <subfield code="a">publication</subfield>
    <subfield code="b">conferencepaper</subfield>
  </datafield>
</record>
61
30
views
downloads
All versions This version
Views 6161
Downloads 3030
Data volume 9.4 MB9.4 MB
Unique views 5353
Unique downloads 3030

Share

Cite as