darknet之VSCode單步調試

1. How

• 參考這篇博客
• launch.json內容如下

  • {
    // Use IntelliSense to learn about possible attributes.
    // Hover to view descriptions of existing attributes.
    // For more information, visit: https://go.microsoft.com/fwlink/?linkid=830387
    "version": "0.2.0",
    "configurations": [
        {
            "name": "region",
            "type": "cppdbg",
            "request": "launch",
            "program": "${workspaceFolder}/darknet",
            "args": ["detector", "train", "${workspaceFolder}/data/candle.data", "${workspaceFolder}/cfg/candle11-tr.cfg", "${workspaceFolder}/yolov2-tiny.conv.11"],
            "stopAtEntry": false,
            "cwd": "${workspaceFolder}",
            "environment": [],
            "externalConsole": false,
            "MIMode": "gdb",
            "preLaunchTask": "make",
            "setupCommands": [
                {
                    "description": "Enable pretty-printing for gdb",
                    "text": "-enable-pretty-printing",
                    "ignoreFailures": true
                }
            ]
        }
    ]}
    

• 修改Makefile，在CFLAGS最後加-g

2. 訓練流程

• darknet.c調用detector.c的run_detector
• run_detector調用train_detector
• train_detector調用parser.c的parse_network_cfg，實際調用parse_network_cfg_custom
  • parse_network_cfg_custom調用解析網絡層的函數，以region層爲例，調用parse_region
  • parse_region調用region_layer.c的make_region_layer，並設置anchor
• train_detector調用data.c的load_data
• train_detector調用network.c的train_network_waitkey（只用1個GPU時）
  • train_network_waitkey調用train_network_datum
  • train_network_datum調用forward_network和backward_network
• train_detector調用validate_detector_map（需要的時候）