Toybrick

标题: 关于RK 输入单通道灰度图片要怎么进行设置呢 [打印本页]

作者: tof3d 时间: 2019-3-6 10:29
标题: 关于RK 输入单通道灰度图片要怎么进行设置呢
本帖最后由 tof3d 于 2019-3-6 12:18 编辑

name: "set003_net"
input: "data"
input_dim: 1
input_dim: 1
input_dim: 128
input_dim: 128

作者: tof3d 时间: 2019-3-6 10:47
本帖最后由 tof3d 于 2019-3-6 10:49 编辑

转换
rknn.config(channel_mean_value='0 0 0 255', reorder_channel='0 1 2')
ret = rknn.load_caffe(model='..C_deploy_new.prototxt',       proto='caffe',       blobs='../C_new.caffemodel')

if ret != 0:
   print('Load mobilenet_v2 failed! Ret = {}'.format(ret))
   exit(ret)
   print('done')

      # Build model

      print('--> Building model')

      print('--> Building model')

      rknn.build(do_quantization=False)

      print('done')

测试
   img = cv2.imread('./ec1.jpg')

#img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

#img=cv2.resize(img,(128,128))

img2 = cv2.imread('./ec2.jpg')

#img2 = cv2.cvtColor(img2, cv2.COLOR_BGR2RGB)

img2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)

img=cv2.resize(img,(128,128))

img2=cv2.resize(img2,(128,128))

print('--> Init runtime environment')

ret = rknn.init_runtime()

if ret != 0:

      print('Init runtime environment failed')

      exit(ret)

print('done')

怎么改，支持单通道吗

# Inference

print('--> Running model')

#outputs = rknn.inference(inputs=[img])

outputs = rknn.inference(inputs=[img])

作者: tof3d 时间: 2019-3-6 11:25
我将rknn.config(channel_mean_value='0 0 0 1‘)，但是输出output全部为0，这个什么问题呢

作者: tof3d 时间: 2019-3-6 12:19
name: "set003_net"
layer {
  name: "input"
  type: "Input"
  top: "data"
  input_param {
shape {
   dim: 1
   dim: 1
   dim: 128
   dim: 128
}
  }
}
layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv1"
  param {
lr_mult: 1
decay_mult: 1
  }
  param {
lr_mult: 2
decay_mult: 0
  }
  convolution_param {
num_output: 96
pad: 2
kernel_size: 5
stride: 1
weight_filler {
   type: "xavier"
}
bias_filler {
   type: "constant"
   value: 0.1
}
  }
}
layer {
  name: "slice1"
  type: "Slice"
  bottom: "conv1"
  top: "slice1_1"
  top: "slice1_2"
  slice_param {
slice_dim: 1
  }
}
layer {
  name: "etlwise1"
  type: "Eltwise"
  bottom: "slice1_1"
  bottom: "slice1_2"
  top: "eltwise1"
  eltwise_param {
operation: MAX
  }
}
layer {
  name: "pool1"
  type: "Pooling"
  bottom: "eltwise1"
  top: "pool1"
  pooling_param {
pool: MAX
kernel_size: 2
stride: 2
  }
}
layer {
  name: "conv2a"
  type: "Convolution"
  bottom: "pool1"
  top: "conv2a"
  param {
lr_mult: 1
decay_mult: 1
  }
  param {
lr_mult: 2
decay_mult: 0
  }
  convolution_param {
num_output: 96
kernel_size: 1
stride: 1
weight_filler {
   type: "xavier"
}
bias_filler {
   type: "constant"
   value: 0.1
}
  }
}
layer {
  name: "slice2a"
  type: "Slice"
  bottom: "conv2a"
  top: "slice2a_1"
  top: "slice2a_2"
  slice_param {
slice_dim: 1
  }
}
layer {
  name: "etlwise2a"
  type: "Eltwise"
  bottom: "slice2a_1"
  bottom: "slice2a_2"
  top: "eltwise2a"
  eltwise_param {
operation: MAX
  }
}
layer {
  name: "conv2"
  type: "Convolution"
  bottom: "eltwise2a"
  top: "conv2"
  param {
lr_mult: 1
decay_mult: 1
  }
  param {
lr_mult: 2
decay_mult: 0
  }
  convolution_param {
num_output: 192
pad: 1
kernel_size: 3
stride: 1
weight_filler {
   type: "xavier"
}
bias_filler {
   type: "constant"
   value: 0.1
}
  }
}
layer {
  name: "slice2"
  type: "Slice"
  bottom: "conv2"
  top: "slice2_1"
  top: "slice2_2"
  slice_param {
slice_dim: 1
  }
}
layer {
  name: "etlwise2"
  type: "Eltwise"
  bottom: "slice2_1"
  bottom: "slice2_2"
  top: "eltwise2"
  eltwise_param {
operation: MAX
  }
}
layer {
  name: "pool2"
  type: "Pooling"
  bottom: "eltwise2"
  top: "pool2"
  pooling_param {
pool: MAX
kernel_size: 2
stride: 2
  }
}
layer {
  name: "conv3a"
  type: "Convolution"
  bottom: "pool2"
  top: "conv3a"
  param {
lr_mult: 1
decay_mult: 1
  }
  param {
lr_mult: 2
decay_mult: 0
  }
  convolution_param {
num_output: 192
kernel_size: 1
stride: 1
weight_filler {
   type: "xavier"
}
bias_filler {
   type: "constant"
   value: 0.1
}
  }
}
layer {
  name: "slice3a"
  type: "Slice"
  bottom: "conv3a"
  top: "slice3a_1"
  top: "slice3a_2"
  slice_param {
slice_dim: 1
  }
}
layer {
  name: "etlwise3a"
  type: "Eltwise"
  bottom: "slice3a_1"
  bottom: "slice3a_2"
  top: "eltwise3a"
  eltwise_param {
operation: MAX
  }
}
layer {
  name: "conv3"
  type: "Convolution"
  bottom: "eltwise3a"
  top: "conv3"
  param {
lr_mult: 1
decay_mult: 1
  }
  param {
lr_mult: 2
decay_mult: 0
  }
  convolution_param {
num_output: 384
pad: 1
kernel_size: 3
stride: 1
weight_filler {
   type: "xavier"
}
bias_filler {
   type: "constant"
   value: 0.1
}
  }
}
layer {
  name: "slice3"
  type: "Slice"
  bottom: "conv3"
  top: "slice3_1"
  top: "slice3_2"
  slice_param {
slice_dim: 1
  }
}
layer {
  name: "etlwise3"
  type: "Eltwise"
  bottom: "slice3_1"
  bottom: "slice3_2"
  top: "eltwise3"
  eltwise_param {
operation: MAX
  }
}
layer {
  name: "pool3"
  type: "Pooling"
  bottom: "eltwise3"
  top: "pool3"
  pooling_param {
pool: MAX
kernel_size: 2
stride: 2
  }
}
layer {
  name: "conv4a"
  type: "Convolution"
  bottom: "pool3"
  top: "conv4a"
  param {
lr_mult: 1
decay_mult: 1
  }
  param {
lr_mult: 2
decay_mult: 0
  }
  convolution_param {
num_output: 384
kernel_size: 1
stride: 1
weight_filler {
   type: "xavier"
}
bias_filler {
   type: "constant"
   value: 0.1
}
  }
}
layer {
  name: "slice4a"
  type: "Slice"
  bottom: "conv4a"
  top: "slice4a_1"
  top: "slice4a_2"
  slice_param {
slice_dim: 1
  }
}
layer {
  name: "etlwise4a"
  type: "Eltwise"
  bottom: "slice4a_1"
  bottom: "slice4a_2"
  top: "eltwise4a"
  eltwise_param {
operation: MAX
  }
}
layer {
  name: "conv4"
  type: "Convolution"
  bottom: "eltwise4a"
  top: "conv4"
  param {
lr_mult: 1
decay_mult: 1
  }
  param {
lr_mult: 2
decay_mult: 0
  }
  convolution_param {
num_output: 256
pad: 1
kernel_size: 3
stride: 1
weight_filler {
   type: "xavier"
}
bias_filler {
   type: "constant"
   value: 0.1
}
  }
}
layer {
  name: "slice4"
  type: "Slice"
  bottom: "conv4"
  top: "slice4_1"
  top: "slice4_2"
  slice_param {
slice_dim: 1
  }
}
layer {
  name: "etlwise4"
  type: "Eltwise"
  bottom: "slice4_1"
  bottom: "slice4_2"
  top: "eltwise4"
  eltwise_param {
operation: MAX
  }
}
layer {
  name: "conv5a"
  type: "Convolution"
  bottom: "eltwise4"
  top: "conv5a"
  param {
lr_mult: 1
decay_mult: 1
  }
  param {
lr_mult: 2
decay_mult: 0
  }
  convolution_param {
num_output: 256
kernel_size: 1
stride: 1
weight_filler {
   type: "xavier"
}
bias_filler {
   type: "constant"
   value: 0.1
}
  }
}
layer {
  name: "slice5a"
  type: "Slice"
  bottom: "conv5a"
  top: "slice5a_1"
  top: "slice5a_2"
  slice_param {
slice_dim: 1
  }
}
layer {
  name: "etlwise5a"
  type: "Eltwise"
  bottom: "slice5a_1"
  bottom: "slice5a_2"
  top: "eltwise5a"
  eltwise_param {
operation: MAX
  }
}
layer {
  name: "conv5"
  type: "Convolution"
  bottom: "eltwise5a"
  top: "conv5"
  param {
lr_mult: 1
decay_mult: 1
  }
  param {
lr_mult: 2
decay_mult: 0
  }
  convolution_param {
num_output: 256
pad: 1
kernel_size: 3
stride: 1
weight_filler {
   type: "xavier"
}
bias_filler {
   type: "constant"
   value: 0.1
}
  }
}
layer {
  name: "slice5"
  type: "Slice"
  bottom: "conv5"
  top: "slice5_1"
  top: "slice5_2"
  slice_param {
slice_dim: 1
  }
}
layer {
  name: "etlwise5"
  type: "Eltwise"
  bottom: "slice5_1"
  bottom: "slice5_2"
  top: "eltwise5"
  eltwise_param {
operation: MAX
  }
}
layer {
  name: "pool4"
  type: "Pooling"
  bottom: "eltwise5"
  top: "pool4"
  pooling_param {
pool: MAX
kernel_size: 2
stride: 2
  }
}
layer {
  name: "fc1"
  type: "InnerProduct"
  bottom: "pool4"
  top: "fc1"
  param {
lr_mult: 1
decay_mult: 1
  }
  param {
lr_mult: 2
decay_mult: 0
  }
  inner_product_param {
num_output: 512
weight_filler {
   type: "xavier"
}
bias_filler {
   type: "constant"
   value: 0.1
}
  }
}
layer {
  name: "slice_fc1"
  type: "Slice"
  bottom: "fc1"
  top: "slice_fc1_1"
  top: "slice_fc1_2"
  slice_param {
slice_dim: 1
  }
}
layer {
  name: "etlwise_fc1"
  type: "Eltwise"
  bottom: "slice_fc1_1"
  bottom: "slice_fc1_2"
  top: "eltwise_fc1"
  eltwise_param {
operation: MAX
  }
}
#layer {
#  name: "drop1"
#  type: "Dropout"
#  bottom: "eltwise_fc1"
#  top: "eltwise_fc1"
#  dropout_param {
# dropout_ratio: 0.7
#  }
#}

作者: seedlin 时间: 2019-3-6 14:55

tof3d 发表于 2019-3-6 12:19
name: "set003_net"
layer {
name: "input"

单通道，你config函数里面参数去掉试试

作者: tof3d 时间: 2019-3-6 15:09

seedlin 发表于 2019-3-6 14:55
单通道，你config函数里面参数去掉试试

img = cv2.imread('./Rec1.jpg',0)

#img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

img=cv2.resize(img,(128,128))

img2=cv2.resize(img2,(128,128))

print('--> Init runtime environment')

ret = rknn.init_runtime()

if ret != 0:

print('Init runtime environment failed')

exit(ret)

print('done')

# Inference

print('--> Running model')

import pdb

pdb.set_trace()

#outputs = rknn.inference(inputs=[img])

img=img.reshape(1,128,128)

outputs = rknn.inference(inputs=[img])
删除以后再rknn.inference 直接Segmentation fault (core dumped)，请问传进去的image 要做shape转换吗，如果不转换也是报这个错

作者: tof3d 时间: 2019-3-6 15:15
如图[attach]160[/attach]

作者: tof3d 时间: 2019-3-6 17:26
help 啊啊

作者: seedlin 时间: 2019-3-6 17:43

tof3d 发表于 2019-3-6 15:09
img = cv2.imread('./Rec1.jpg',0)

#img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

怎么么看到rknn load_model的代码

作者: tof3d 时间: 2019-3-6 18:53

seedlin 发表于 2019-3-6 17:43
怎么么看到rknn load_model的代码

没有贴出来
import numpy as np
import cv2
from rknn.api import RKNN
from scipy import spatial

if __name__ == '__main__':

# Create RKNN object

rknn = RKNN()

rknn = RKNN(verbose=True)

# pre-process config

print('--> config model')

#rknn.config(channel_mean_value='103.94 116.78 123.68 58.82', reorder_channel='2 1 0')

print('done')

# Load tensorflow model

print('--> Loading model')

if(1):

#  只在屏幕打印详细的日志信息

      rknn = RKNN(verbose=True)

      #rknn.config(channel_mean_value='0 0 0 255', reorder_channel='0 1 2')

      #rknn.config(channel_mean_value='0 0 0 1')

      ret = rknn.load_caffe(model='../C_deploy_new.prototxt',       proto='caffe',       blobs='../C_new.caffemodel')

      if ret != 0:

         print('Load mobilenet_v2 failed! Ret = {}'.format(ret))

         exit(ret)

      print('done')

      # Build model

      print('--> Building model')

      print('--> Building model')

      rknn.build(do_quantization=False)

      print('done')

# Export rknn model

      print('--> Export RKNN model')

      ret = rknn.export_rknn('../Mobile-ligth.rknn')

      if ret != 0:

         print('Export mobilenet_v2.rknn failed!')

         exit(ret)

      print('done')

ret = rknn.load_rknn(path='../Mobile-ligth.rknn')

#ret = rknn.init_runtime(target='rk3399pro')

# Set inputs

img = cv2.imread('./ec1.jpg')

img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

img = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)

#img=cv2.resize(img,(128,128))

img2 = cv2.imread('./ec2.jpg')

#img2 = cv2.cvtColor(img2, cv2.COLOR_BGR2RGB)

img2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)

img = cv2.imread('./ec1.jpg',0)

#img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

img=cv2.resize(img,(128,128))

img2=cv2.resize(img2,(128,128))

print('--> Init runtime environment')

ret = rknn.init_runtime()

if ret != 0:

      print('Init runtime environment failed')

      exit(ret)

print('done')

# Inference

print('--> Running model')

import pdb

pdb.set_trace()

#outputs = rknn.inference(inputs=[img])

img=img.reshape(1,128,128)

outputs = rknn.inference(inputs=[img])

if(1):

      import pdb

      pdb.set_trace()

      outputs2 = rknn.inference(inputs=[img2])

      print(outputs,outputs2[0])

      import pdb

      #pdb.set_trace()

      result = 1 - spatial.distance.cosine(outputs, outputs2)

      print(result)

#show_outputs(outputs)

print('done')

# perf

print('--> Begin evaluate model performance')

#perf_results = rknn.eval_perf(inputs=[img])

print('done')

rknn.release()

作者: seedlin 时间: 2019-3-7 09:16

tof3d 发表于 2019-3-6 18:53
没有贴出来
import numpy as np
import cv2

RKNN对象不是只创建一次就可以了，你为啥重复创建了几次

作者: jefferyzhang 时间: 2019-3-7 09:35
车牌识别就是灰度图，单通道输入的。
可以参考下：http://t.rock-chips.com/forum.php?mod=viewthread&tid=98

作者: tof3d 时间: 2019-3-7 10:47

jefferyzhang 发表于 2019-3-7 09:35
车牌识别就是灰度图，单通道输入的。
可以参考下：http://t.rock-chips.com/forum.php?mod=viewthread&tid= ...

大佬，你这个tensorflow 模型啊，好像按照他的改了，效果不大啊，可以帮我看看我的模型吗，链接：https://pan.baidu.com/s/1NBfVQB5Fo9xFr7MehmT3MQ
提取码：erf9

作者: jefferyzhang 时间: 2019-3-7 12:58

tof3d 发表于 2019-3-7 10:47
大佬，你这个tensorflow 模型啊，好像按照他的改了，效果不大啊，可以帮我看看我的模型吗，链接：https:/ ...

可我不会caffe啊，我建议你就先写一层max pool，然后输出，这样就可以对下问题了

作者: tof3d 时间: 2019-3-7 14:00

jefferyzhang 发表于 2019-3-7 12:58
可我不会caffe啊，我建议你就先写一层max pool，然后输出，这样就可以对下问题了 ...

我测试了
name: "DeepFace_set003_net"
layer {
  name: "input"
  type: "Input"
  top: "data"
  input_param {
shape {
   dim: 1
   dim: 1
   dim: 128
   dim: 128
}
  }
}
layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv1"
  param {
lr_mult: 1
decay_mult: 1
  }
  param {
lr_mult: 2
decay_mult: 0
  }
  convolution_param {
num_output: 96
pad: 2
kernel_size: 5
stride: 1
weight_filler {
   type: "xavier"
}
bias_filler {
   type: "constant"
   value: 0.1
}
  }
}
layer {
  name: "slice1"
  type: "Slice"
  bottom: "conv1"
  top: "slice1_1"
  top: "slice1_2"
  slice_param {
slice_dim: 1
  }
}
layer {
  name: "etlwise1"
  type: "Eltwise"
  bottom: "slice1_1"
  bottom: "slice1_2"
  top: "eltwise1"
  eltwise_param {
operation: MAX
  }
}
这几层可以用，output有结果

作者: jefferyzhang 时间: 2019-3-7 14:22

tof3d 发表于 2019-3-7 14:00
我测试了
name: "DeepFace_set003_net"
layer {

那就好，至少排除了灰度图输入出现问题的可能。

作者: tof3d 时间: 2019-3-7 14:32

jefferyzhang 发表于 2019-3-7 14:22
那就好，至少排除了灰度图输入出现问题的可能。

我发现当我把 rknn.config 设置为 rknn.config(channel_mean_value='0 0 0 255', reorder_channel='0 1 2')，不会出现segment 错误，但是结果全部是0，为什么单通道不行呢

作者: tof3d 时间: 2019-3-7 14:37
这边还有这样的warning
W [vxeltwiseMaxValidator:132][src/libnnext/ops/kernel/RK_nn_kernel_eltwisemax.c : 132] Warning num_of_dim<2 || num_of_dim > 3

作者: tof3d 时间: 2019-3-7 16:55

等待大神帮助

作者: tof3d 时间: 2019-3-8 11:15

，pending for help，我怀疑还是输入的问题，要怎么搞呢

作者: tof3d 时间: 2019-3-11 10:29
没法解决吗

作者: tof3d 时间: 2019-3-11 20:41

作者: tof3d 时间: 2019-3-14 11:06
help me please ,the relative models and files can be downloaded from baidu disk,here is the link:
link：https://pan.baidu.com/s/1nBNkiyekpmdpEEH0tVF7EA
pwd：vdm7

作者: Lee 时间: 2019-3-29 16:57

tof3d 发表于 2019-3-14 11:06
help me please ,the relative models and files can be downloaded from baidu disk,here is the link:
...

我遇到了这个问题，你解决啦吗，难道只有三通道才不会有问题吗？

欢迎光临 Toybrick (https://t.rock-chips.com/)