Faster-RCNN+VGG用自己的数据集训练模型

和ZF差不多，基本一样。不同的地方主要是网络模型的修改和训练结束后的修改。

1-6参考Faster-RCNN+ZF用自己的数据集训练模型

7.模型的修改

（1）models\fast_rcnn_prototxts\vgg_16layers_fc6\train_val.prototxt

input: "bbox_targets"
input_dim: 1  # to be changed on-the-fly to match num ROIs
input_dim: 20 #  4*(类别数+1) (=21) 
input_dim: 1
input_dim: 1

input: "bbox_loss_weights"
input_dim: 1  # to be changed on-the-fly to match num ROIs
input_dim: 20 # 4*(类别数+1)
input_dim: 1
input_dim: 1

layer {
	bottom: "fc7"
	top: "cls_score"
	name: "cls_score"
	param {
		lr_mult: 1.0
	}
	param {
		lr_mult: 2.0
	}
	type: "InnerProduct"
	inner_product_param {
		num_output: 5 #类别数+1
		weight_filler {
			type: "gaussian"
			std: 0.01
		}
		bias_filler {
			type: "constant"
			value: 0
		}
	}
}

layer {
	bottom: "fc7"
	top: "bbox_pred"
	name: "bbox_pred"
	type: "InnerProduct"
	param {
		lr_mult: 1.0
	}
	param {
		lr_mult: 2.0
	}
	inner_product_param {
		num_output: 20 #4*(类别数+1)
		weight_filler {
			type: "gaussian"
			std: 0.001
		}
		bias_filler {
			type: "constant"
			value: 0
		}
	}
}

（2）models\fast_rcnn_prototxts\vgg_16layers_fc6\test.prototxt

layer {
	bottom: "fc7"
	top: "cls_score"
	name: "cls_score"
	param {
		lr_mult: 1.0
	}
	param {
		lr_mult: 2.0
	}
	type: "InnerProduct"
	inner_product_param {
		num_output: 5 #类别数+1
		weight_filler {
			type: "gaussian"
			std: 0.01
		}
		bias_filler {
			type: "constant"
			value: 0
		}
	}
}

layer {
	bottom: "fc7"
	top: "bbox_pred"
	name: "bbox_pred"
	type: "InnerProduct"
	param {
		lr_mult: 1.0
	}
	param {
		lr_mult: 2.0
	}
	inner_product_param {
		num_output: 20 #4*(类别数+1)
		weight_filler {
			type: "gaussian"
			std: 0.001
		}
		bias_filler {
			type: "constant"
			value: 0
		}
	}
}

（3）models\fast_rcnn_prototxts\vgg_16layers_conv3_1\train_val.prototxt

input: "bbox_targets"
input_dim: 1  # to be changed on-the-fly to match num ROIs
input_dim: 20 # 4*(类别数+1) 
input_dim: 1
input_dim: 1

input: "bbox_loss_weights"
input_dim: 1  # to be changed on-the-fly to match num ROIs
input_dim: 20 # 4* (类别数+1) 
input_dim: 1
input_dim: 1

layer {
	bottom: "fc7"
	top: "cls_score"
	name: "cls_score"
	param {
		lr_mult: 1.0
	}
	param {
		lr_mult: 2.0
	}
	type: "InnerProduct"
	inner_product_param {
		num_output: 5 #类别数+1
		weight_filler {
			type: "gaussian"
			std: 0.01
		}
		bias_filler {
			type: "constant"
			value: 0
		}
	}
}

layer {
	bottom: "fc7"
	top: "bbox_pred"
	name: "bbox_pred"
	type: "InnerProduct"
	param {
		lr_mult: 1.0
	}
	param {
		lr_mult: 2.0
	}
	inner_product_param {
		num_output: 20 #4*(类别数+1)
		weight_filler {
			type: "gaussian"
			std: 0.001
		}
		bias_filler {
			type: "constant"
			value: 0
		}
	}
}

（4）models\fast_rcnn_prototxts\vgg_16layers_conv3_1\test.prototxt

layer {
	bottom: "fc7"
	top: "cls_score"
	name: "cls_score"
	param {
		lr_mult: 1.0
	}
	param {
		lr_mult: 2.0
	}
	type: "InnerProduct"
	inner_product_param {
		num_output:5  #类别数+1
		weight_filler {
			type: "gaussian"
			std: 0.01
		}
		bias_filler {
			type: "constant"
			value: 0
		}
	}
}

layer {
	bottom: "fc7"
	top: "bbox_pred"
	name: "bbox_pred"
	type: "InnerProduct"
	param {
		lr_mult: 1.0
	}
	param {
		lr_mult: 2.0
	}
	inner_product_param {
		num_output: 20 #4*(类别数+1)
		weight_filler {
			type: "gaussian"
			std: 0.001
		}
		bias_filler {
			type: "constant"
			value: 0
		}
	}
}

！！！为防止与之前的模型搞混,训练前把output文件夹删除（或改个其他名），还要把imdb\cache中的文件删除（如果有的话）

8.开始训练

experiments\script_faster_rcnn_VOC2007_VGG16.m

9.训练完后

将relu5（包括relu5）前的层删除，并将roi_pool5的bottom改为data和rois。并且前面的input_dim:分别改为1,512,50,50，具体如下：

input: "data"
input_dim: 1
input_dim: 512
input_dim: 50
input_dim: 50

layer {
	bottom: "data"
	bottom: "rois"
	top: "pool5"
	name: "roi_pool5"
	type: "ROIPooling"
	roi_pooling_param {
		pooled_w: 7
		pooled_h: 7
		spatial_scale: 0.0625  # (1/16)
	}
}