• Chinese Journal of Lasers
  • Vol. 51, Issue 3, 0307108 (2024)
Zheng Zhang1, Mingxiao Chen1, Xinyu Li1, Yi Chen1, Shuwei Shen2, and Peng Yao3、aff***
Author Affiliations
  • 1Department of Precision Machinery and Precision Instrumentation, School of Engineering Science, University of Science and Technology of China, Hefei 230027, Anhui , China
  • 2Suzhou Advanced Research Institute, University of Science and Technology of China, Suzhou 215123, Jiangsu ,China
  • 3School of Microelectronics, University of Science and Technology of China, Hefei 230027, Anhui , China
  • show less
    DOI: 10.3788/CJL231261 Cite this Article Set citation alerts
    Zheng Zhang, Mingxiao Chen, Xinyu Li, Yi Chen, Shuwei Shen, Peng Yao. Automatic Identification of Cervical Abnormal Cells Based on Transformer[J]. Chinese Journal of Lasers, 2024, 51(3): 0307108 Copy Citation Text show less
    Overall architecture diagram of model
    Fig. 1. Overall architecture diagram of model
    Architecture diagrams of transformer encoder block. (a) Generic structure; (b) improved structure
    Fig. 2. Architecture diagrams of transformer encoder block. (a) Generic structure; (b) improved structure
    DW convolution diagram
    Fig. 3. DW convolution diagram
    Changes of dynamic IOU thresholds
    Fig. 4. Changes of dynamic IOU thresholds
    Comparison of detection effects of different models. (a) Ground truth; (b) proposed method; (c) Sparse R-CNN
    Fig. 5. Comparison of detection effects of different models. (a) Ground truth; (b) proposed method; (c) Sparse R-CNN
    Loss and AP under dynamic IOU threshold and fixed IOU threshold. (a) Loss; (b) AP
    Fig. 6. Loss and AP under dynamic IOU threshold and fixed IOU threshold. (a) Loss; (b) AP
    Ablation experiment heatmaps. (a) Original images; (b) heatmap generated by original Transformer model; (c) heatmap generated by our method
    Fig. 7. Ablation experiment heatmaps. (a) Original images; (b) heatmap generated by original Transformer model; (c) heatmap generated by our method
    ModelAPAP50AP75APSAPMAPLNumber of parameters /106
    YOLOv318.035.416.80.06.325.961.5
    RetinaNet2823.543.722.50.310.632.237.9
    Sparse R-CNN3224.042.123.20.210.833.4108.5
    Cascade R-CNN3322.139.321.40.27.632.669.3
    DETR2923.546.321.60.59.232.641.5
    FCOS3023.141.022.50.212.232.432.1
    GiraffeDet3123.944.323.10.310.932.647.8
    Iter Sparse R-CNN3424.843.825.30.711.733.0123.4
    Ours26.146.825.51.813.533.548.3
    Table 1. Experimental results of various models
    ModelAPAP50AP75
    attFPN25.050.322.2
    Proposed model26.146.825.5
    Table 2. Comparison between proposed model and attFPN
    BackboneAPAP50AP75APSAPMAPL
    CNN(Resnet-101)23.742.124.21.112.831.4
    Transformer26.146.825.51.813.533.5
    Table 3. Comparison of experimental results under different backbone choices
    Improved Transformer encoderDynamic IOU thresholdAPAP50AP75APSAPMAPL
    24.143.823.80.513.230.7
    24.744.725.20.212.032.2
    25.946.125.32.512.233.8
    26.146.825.51.813.533.5
    Table 4. Ablation experiment results of different modules
    IOU thresholdAPAP50AP75APSAPMAPL
    0.523.645.522.40.412.631.0
    0.624.042.123.20.210.833.4
    0.724.143.823.80.513.230.7
    Dynamic threshold24.744.725.20.212.032.2
    Table 5. Comparison among models based on dynamic IOU threshold and multiple fixed IOU thresholds
    Zheng Zhang, Mingxiao Chen, Xinyu Li, Yi Chen, Shuwei Shen, Peng Yao. Automatic Identification of Cervical Abnormal Cells Based on Transformer[J]. Chinese Journal of Lasers, 2024, 51(3): 0307108
    Download Citation