yolov4

Model descriptions

  • YOLOv4

    • The model in YOLOv4 paper.
    • Backbone - CSPDarknet53 with Mish activation
    • Neck - PANet with Leaky activation
    • Plugin Modules - SPP
    • V100 FPS - 62@608x608, 83@512x512
    • BFLOPs - 128.5@608x608
  • YOLOv4-Leaky

    • Backbone - CSPDarknet53 with Leaky activation
    • Neck - PANet with Leaky activation
    • Plugin Modules - SPP
    • BFLOPs - 128.5@608x608
  • YOLOv4-SAM-Leaky

    • Backbone - CSPDarknet53 with Leaky activation
    • Neck - PANet with Leaky activation
    • Plugin Modules - SPP, SAM(Spatial Attention Module)空间注意模块
    • BFLOPs - 130.7@608x608
  • YOLOv4-Mish

    • Backbone - CSPDarknet53 with Mish activation
    • Neck - PANet with Mish activation
    • Plugin Modules - SPP
    • BFLOPs - 128.5@608x608
  • YOLOv4-SAM-Mish

    • Backbone - CSPDarknet53 with Mish activation
    • Neck - PANet with Mish activation
    • Plugin Modules - SPP, SAM
    • V100 FPS - 61@608x608, 81@512x512
    • BFLOPs - 130.7@608x608
  • YOLOv4-CSP

    • will update
    • Backbone - CSPDarknet53 with Mish activation
    • Neck - CSPPANet with Mish activation
    • Plugin Modules - SPP
  • YOLOv4-CSP-SAM

    • will update
    • Backbone - CSPDarknet53 with Mish activation
    • Neck - CSPPANet with Mish activation
    • Plugin Modules - SPP, SAM

你可能感兴趣的:(yolo系列)