YOLO3网络层具体参数的输出位置

  1. 卷积层的输出

0 conv     32       3 x 3/ 1    416 x 416 x   3 ->  416 x 416 x  32 0.299 BF

输出位置:

darknet-master\src\convolutional_layer.c(564):    fprintf(stderr, "%4d x%4d x%4d -> %4d x%4d x%4d %5.3f BF\n", w, h, c, l.out_w, l.out_h, l.out_c, l.bflops);

2.Crop Layer的输出位置

C:\darknet-master\src\crop_layer.c(18):    fprintf(stderr, "Crop Layer: %d x %d -> %d x %d x %d image\n", h,w,crop_height,crop_width,c);

3.反卷积层的输出

  C:\darknet-master\src\deconvolutional_layer.c(101):    fprintf(stderr, "Deconvolutional Layer: %d x %d x %d image, %d filters -> %d x %d x %d image\n", h,w,c,n, out_h, out_w, n);

4.Local Layer

C:\darknet-master\src\local_layer.c(86):    fprintf(stderr, "Local Layer: %d x %d x %d image, %d filters -> %d x %d x %d image\n", h,w,c,n, out_h, out_w, n);

5.LSTM Layer

C:\darknet-master\src\lstm_layer.c(31):    fprintf(stderr, "LSTM Layer: %d inputs, %d outputs\n", inputs, outputs);
  1. 最大池化层的输出

C:\darknet-master\src\maxpool_layer.c(90):    fprintf(stderr, "max               %d x %d/%2d   %4d x%4d x%4d -> %4d x%4d x%4d %5.3f BF\n", size, size, stride, w, h, c, l.out_w, l.out_h, l.out_c, l.bflops);

7.reorg

 C:\darknet-master\src\reorg_layer.c(26):    fprintf(stderr, "reorg                    /%2d %4d x%4d x%4d -> %4d x%4d x%4d\n",  stride, w, h, c, l.out_w, l.out_h, l.out_c);

8.downsample 与upsample

C:\darknet-master\src\upsample_layer.c(39):    if(l.reverse) fprintf(stderr, "downsample              %2dx  %4d x%4d x%4d -> %4d x%4d x%4d\n", stride, w, h, c, l.out_w, l.out_h, l.out_c);
  C:\darknet-master\src\upsample_layer.c(40):    else fprintf(stderr, "upsample                %2dx  %4d x%4d x%4d -> %4d x%4d x%4d\n", stride, w, h, c, l.out_w, l.out_h, l.out_c);
  1. yolo层的输出

82 yolo
C:\darknet-master\src\yolo_layer.c(72):    fprintf(stderr, "yolo\n");
  1. Shortcut Layer的输出

4 Shortcut Layer: 1
C:\darknet-master\src\shortcut_layer.c(9):    fprintf(stderr,"Shortcut Layer: %d\n", index);
  1. route

83 route  79
C:\darknet-master\src\route_layer.c(8):    fprintf(stderr,"route ");
  C:\darknet-master\src\route_layer.c(18):        fprintf(stderr," %d", input_layers[i]);
  C:\darknet-master\src\route_layer.c(21):    fprintf(stderr, "\n");
  1. 抬头

C:\darknet-master\src\parser.c(881):    fprintf(stderr, "   layer   filters  size/strd(dil)      input                output\n");
  1. 每行开头的序号

C:\darknet-master\src\parser.c(892):    fprintf(stderr, "%4d ", count);

13.分配内存

C:\darknet-master\src\parser.c(1021):            fprintf(stderr, " Allocate additional workspace_size = %1.2f MB \n", (float)workspace_size/1000000);
  1. 综合信息

C:\darknet-master\src\parser.c(358):    fprintf(stderr, "[yolo] params: iou loss: %s, iou_norm: %2.2f, cls_norm: %2.2f, scale_x_y: %2.2f\n", (l.iou_loss == MSE ? "mse" : (l.iou_loss == GIOU ? "giou" : "iou")), l.iou_normalizer, l.cls_normalizer, l.scale_x_y);
  1. try to allocate additional workspace_size

C:\darknet-master\src\network.c(446):        printf("\n try to allocate additional workspace_size = %1.2f MB \n", (float)workspace_size / 1000000);
  C:\darknet-master\src\network.c(448):        printf(" CUDA allocate done! \n");
  C:\darknet-master\src\network.c(547):            fprintf(stderr, "Resizing type %d \n", (int)l.type);
  C:\darknet-master\src\network.c(560):        printf(" try to allocate additional workspace_size = %1.2f MB \n", (float)workspace_size / 1000000);
  C:\darknet-master\src\network.c(570):        printf(" CUDA allocate done! \n");
  1. Total BFLOPS 的输出

 C:\darknet-master\src\parser.c(999):    fprintf(stderr, "Total BFLOPS %5.3f \n", bflops);
  1. seen 64的显示

C:\darknet-master\src\parser.c(1394):        printf("\n seen 64 \n");
  1. Done!的显示

C:\darknet-master\src\parser.c(1475):    fprintf(stderr, "Done!\n");
  1. compute_capability的输出

C:\darknet-master\src\parser.c(782):        fprintf(stderr, " compute_capability = %d, cudnn_half = %d \n", compute_capability, net->cudnn_half);

你可能感兴趣的:(darknet)