Laser & Optoelectronics Progress, Volume. 61, Issue 10, 1037006(2024)

Efficient Global Attention Networks for Image Super-Resolution Reconstruction

Qingqing Wang1,2, Yuelan Xin1,2、*, Jia Zhao2, Jiang Guo1,2, and Haochen Wang2
Author Affiliations
  • 1The College of Computer, Qinghai Normal University, Xining 810001, Qinghai, China
  • 2The State Key Laboratory of Tibetan Intelligent Information Processing and Application, Xining 810001, Qinghai, China
  • show less
    Figures & Tables(14)
    Network architecture. (a) EGAN; (b) CAFB; (c) CCA; (d) PA; (e) NPRB
    Affinity operation
    Aggregation operation
    Visualization results of multistage dynamic cosine thermal restart training strategy
    Comparison of model complexity and performance of EGAN proposed in this study with other methods on the BSD100 dataset based on ×4SR. The size of the circle indicates the number of parameters
    Comparison of visualization results with advanced algorithms
    • Table 1. Test set information

      View table

      Table 1. Test set information

      DatasetNumber of imagesContent scene
      Set5295Face,bird,butterfly
      Set143014People,plants and animals,natural scenery,PPT,etc.
      BSD10031100People,plants and animals,and indoor and outdoor environments,natural scenery,etc.
      Urban10032100Cityscape
      Manga10933109Comics,backgrounds,etc.
    • Table 2. Parameter setting for network training

      View table

      Table 2. Parameter setting for network training

      Training programParameterization
      OptimizerAdam
      ε21×10-6
      rinitial_learning0.01
      ηmax0.1
      T0100
      Batchsize16
      Epoch500
    • Table 3. Quantitative comparison results of each algorithm on five benchmark test sets

      View table

      Table 3. Quantitative comparison results of each algorithm on five benchmark test sets

      ScaleMethodParameters /103FLOPs /109Set5Set14BSD100Urban100Manga109
      PSNR /dB(SSIM)PSNR /dB(SSIM)PSNR /dB(SSIM)PSNR /dB(SSIM)PSNR /dB(SSIM)
      ×2Bicubic33.66(0.9299)30.24(0.8688)29.56(0.8431)26.88(0.8403)30.80(0.9399)
      VDSR466561337.53(0.9587)33.03(0.9124)31.90(0.8960)30.76(0.9140)37.22(0.9729)
      IDN1055312537.83(0.9600)33.30(0.9148)32.08(0.8985)31.27(0.9196)38.01(0.9749)
      CARN11159222337.76(0.9590)33.52(0.9166)32.09(0.8978)31.92(0.9256)38.36(0.9765)
      IMDN1269415938.00(0.9605)33.63(0.9177)32.19(0.8996)32.17(0.9283)38.01(0.9749)
      SMSR1698513238.00(0.9601)33.64(0.9179)32.17(0.8990)32.19(0.9284)38.76(0.9771)
      LAGNet134478437.79(0.9594)33.40(0.9162)32.10(0.8991)
      LMDFFN1433937.92(0.9602)33.43(0.9163)32.08(0.8985)31.85(0.9250)38.52(0.9766)
      RCCN938.13(0.9610)33.67(0.9185)32.22(0.9001)32.41(0.9302)
      ShuffleMixer153949138.01(0.9606)33.63(0.9180)32.17(0.8995)31.89(0.9257)38.83(0.9774)
      DLGSANet174730109738.34(0.9617)34.25(0.9231)32.38(0.9025)33.41(0.9393)39.57(0.9789)
      HIAAN1863114538.04(0.9611)33.55(0.9186)32.19(0.9006)32.24(0.9293)38.65(0.9773)
      EGAN(Ours)3042038.19(0.9613)34.40(0.9260)34.00(0.9255)32.92(0.9262)38.70(0.9777)
      ×3Bicubic30.39(0.8682)27.55(0.7742)27.21(0.7385)24.46(0.7349)26.95(0.8556)
      VDSR466561333.66(0.9213)29.77(0.8314)28.82(0.7976)27.14(0.8279)32.01(0.9310)
      IDN105535634.11(0.9253)29.99(0.8354)28.95(0.8013)27.42(0.8359)32.71(0.9381)
      CARN11159211934.29(0.9255)30.29(0.8407)29.06(0.8034)28.06(0.8493)33.50(0.9440)
      IMDN127037234.36(0.9270)30.32(0.8417)29.09(0.8046)28.17(0.8519)33.61(0.9445)
      SMSR169936834.40(0.9270)30.33(0.8412)29.10(0.8050)28.25(0.8536)33.68(0.9445)
      LAGNet134567534.26(0.9253)30.22(0.8421)28.93(0.8024)
      LMDFFN1434134.32(0.9264)30.20(0.8392)29.03(0.8034)28.01(0.8483)33.36(0.9430)
      RCCN934.48(0.9278)30.37(0.8417)29.12(0.8054)28.34(0.8549)
      ShuffleMixer154154334.40(0.9272)30.37(0.8423)29.12(0.8051)28.08(0.8498)33.69(0.9448)
      DLGSANet17474048634.95(0.9310)30.77(0.8501)29.38(0.8121)29.43(0.8761)34.76(0.9517)
      HIAAN186406534.44(0.9278)30.30(0.8421)29.13(0.8074)28.27(0.8544)33.64(0.9453)
      EGAN(Ours)310934.97(0.9396)31.01(0.8577)29.60(0.8190)29.21(0.8713)33.81(0.9481)
      ×4Bicubic28.42(0.8104)26.00(0.7027)25.96(0.6675)23.14(0.6577)24.89(0.7866)
      VDSR466561331.35(0.8838)28.01(0.7674)27.29(0.7251)25.18(0.7524)28.83(0.8809)
      IDN105533231.82(0.8903)28.25(0.7730)27.41(0.7297)25.41(0.7632)29.41(0.8942)
      CARN1115929132.13(0.8937)28.60(0.7806)27.58(0.7349)26.07(0.7837)30.47(0.9084)
      IMDN127154132.11(0.8934)28.63(0.7823)27.58(0.7358)26.10(0.7846)30.55(0.9072)
      SMSR1610064232.12(0.8932)28.55(0.7808)27.55(0.7351)26.11(0.7868)30.54(0.9085)
      LAGNet134706832.06(0.8912)28.47(0.7782)27.54(0.7347)
      LMDFFN1434432.08(0.8930)28.46(0.7792)27.51(0.7341)25.93(0.7804)30.25(0.9053)
      RCCN932.24(0.8956)28.69(0.7833)27.63(0.7376)26.31(0.7912)
      ShuffleMixer154112832.21(0.8953)28.66(0.7827)27.61(0.7366)26.08(0.7835)30.65(0.9093)
      DLGSANet17476027432.80(0.9021)28.95(0.7907)27.85(0.7464)27.17(0.8175)31.68(0.9219)
      HIAAN186523732.26(0.8964)28.62(0.7832)27.59(0.7386)26.17(0.7886)30.54(0.9095)
      EGAN(Ours)312832.35(0.8965)28.97(0.7910)28.47(0.7642)26.04(0.7837)30.74(0.9145)
    • Table 4. Effect of different components in CAFB on model performance

      View table

      Table 4. Effect of different components in CAFB on model performance

      CCA24AWANumber of parameters /103PSNR /dB(SSIM)
      28925.89(0.7831)
      30525.96(0.7834)
      31226.04(0.7837)
    • Table 5. Effect of different numbers of CAFBs on model performance

      View table

      Table 5. Effect of different numbers of CAFBs on model performance

      Number of CAFBsNumber of parameters /103PSNR /dB(SSIM)
      431125.43(0.7691)
      631125.73(0.7745)
      831225.69(0.7762)
      1031226.02(0.7820)
      1231226.04(0.7837)
      1431225.62(0.7723)
    • Table 6. Effect of NPRB on HR image reconstruction performance

      View table

      Table 6. Effect of NPRB on HR image reconstruction performance

      NPRBPixel ShuffleNumber of parameters /103PSNR /dB(SSIM)
      30424.69(0.6868)
      31226.04(0.7837)
    • Table 7. Effect of thermal restart strategy on model performance

      View table

      Table 7. Effect of thermal restart strategy on model performance

      MethodPSNR /dB(SSIM)
      Set5Set14BSD100Urban100Manga109
      EGAN-Lr31.91(0.8913)28.41(0.7778)28.33(0.7604)25.73(0.7726)30.26(0.9088)
      EGAN-DCwr32.35(0.8965)28.97(0.7910)28.47(0.7642)26.04(0.7837)30.74(0.9145)
    • Table 8. Effect of the thermal restart strategy on IMDN performance

      View table

      Table 8. Effect of the thermal restart strategy on IMDN performance

      MethodPSNR /dB(SSIM)
      Set5Set14BSD100Urban100Manga109
      IMDN1232.21(0.8948)28.58(0.7811)27.56(0.7353)26.04(0.7838)30.45(0.9075)
      IMDN-DCwr32.23(0.8948)28.58(0.7813)27.56(0.7354)26.08(0.7859)30.47(0.9077)
    Tools

    Get Citation

    Copy Citation Text

    Qingqing Wang, Yuelan Xin, Jia Zhao, Jiang Guo, Haochen Wang. Efficient Global Attention Networks for Image Super-Resolution Reconstruction[J]. Laser & Optoelectronics Progress, 2024, 61(10): 1037006

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Digital Image Processing

    Received: Sep. 5, 2023

    Accepted: Oct. 20, 2023

    Published Online: May. 9, 2024

    The Author Email: Yuelan Xin (xinyue001112@163.com)

    DOI:10.3788/LOP232053

    CSTR:32186.14.LOP232053

    Topics