caffe输出目标检测指标map

修改solver.prototxt文件中的参数：

test_iter:    #设置成测试图片数量除以测试的batchsize

max_iter: 0  #这样就不用训练直接进行测试了

show_per_class_result:true  #统计每个类的mAP值

 

示例：

train_net: "models/VGGNet/mydata/SSD_300x300/train.prototxt"
test_net: "models/VGGNet/mydata/SSD_300x300/test.prototxt"
test_iter: 189
test_interval: 5000
base_lr: 0.001
display: 10
max_iter: 0
lr_policy: "multistep"
gamma: 0.1
momentum: 0.9
weight_decay: 0.0005
snapshot: 2000
snapshot_prefix: "models/VGGNet/mydata/pretrained"
solver_mode: GPU
device_id: 0
debug_info: false
snapshot_after_train: true
test_initialization: false
average_loss: 10
stepvalue: 5000
stepvalue: 10000
stepvalue: 20000
stepvalue: 30000
iter_size: 1
type: "SGD"
eval_type: "detection"
ap_version: "11point"
show_per_class_result:true

和训练同样的操作（我是通过sh文件启动的）

输出：