ncnnoptimize yolov3-tiny error

infrontofme commented 4 years ago

我想把darknet框架下yolov3-tiny的模型转到ncnn 原始yolov3-tiny.cfg文件如下：

[net]
# Testing
batch=1
subdivisions=1
# Training
# batch=64
# subdivisions=2
width=416
height=416
channels=3
momentum=0.9
decay=0.0005
angle=0
saturation = 1.5
exposure = 1.5
hue=.1

learning_rate=0.001
burn_in=1000
max_batches = 500200
policy=steps
steps=400000,450000
scales=.1,.1

[convolutional]
batch_normalize=1
filters=16
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=32
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=1

[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky

###########

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[convolutional]
size=1
stride=1
pad=1
filters=255
activation=linear

[yolo]
mask = 3,4,5
anchors = 10,14,  23,27,  37,58,  81,82,  135,169,  344,319
classes=80
num=6
jitter=.3
ignore_thresh = .7
truth_thresh = 1
random=1

[route]
layers = -4

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[upsample]
stride=2

[route]
layers = -1, 8

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[convolutional]
size=1
stride=1
pad=1
filters=255
activation=linear

[yolo]
mask = 0,1,2
anchors = 10,14,  23,27,  37,58,  81,82,  135,169,  344,319
classes=80
num=6
jitter=.3
ignore_thresh = .7
truth_thresh = 1
random=1

step1：darknet2ncnn，成功生成了.param和.bin文件，.param文件内容如下：

7767517
47 49
Input                  data                 0 1 data 0=416 1=416 2=3
Convolution            0_27                 1 1 data 0_27 0=16 1=3 2=1 3=1 4=1 5=0 6=432
BatchNorm              0_27_bn              1 1 0_27 0_27_bn 0=16 1=.00001
ReLU                   0_27_bn_leaky        1 1 0_27_bn 0_27_bn_leaky 0=0.1
Pooling                1_35                 1 1 0_27_bn_leaky 1_35 0=0 1=2 2=2 5=1 3=0 13=0 14=0 15=0
Convolution            2_39                 1 1 1_35 2_39 0=32 1=3 2=1 3=1 4=1 5=0 6=4608
BatchNorm              2_39_bn              1 1 2_39 2_39_bn 0=32 1=.00001
ReLU                   2_39_bn_leaky        1 1 2_39_bn 2_39_bn_leaky 0=0.1
Pooling                3_47                 1 1 2_39_bn_leaky 3_47 0=0 1=2 2=2 5=1 3=0 13=0 14=0 15=0
Convolution            4_51                 1 1 3_47 4_51 0=64 1=3 2=1 3=1 4=1 5=0 6=18432
BatchNorm              4_51_bn              1 1 4_51 4_51_bn 0=64 1=.00001
ReLU                   4_51_bn_leaky        1 1 4_51_bn 4_51_bn_leaky 0=0.1
Pooling                5_59                 1 1 4_51_bn_leaky 5_59 0=0 1=2 2=2 5=1 3=0 13=0 14=0 15=0
Convolution            6_63                 1 1 5_59 6_63 0=128 1=3 2=1 3=1 4=1 5=0 6=73728
BatchNorm              6_63_bn              1 1 6_63 6_63_bn 0=128 1=.00001
ReLU                   6_63_bn_leaky        1 1 6_63_bn 6_63_bn_leaky 0=0.1
Pooling                7_71                 1 1 6_63_bn_leaky 7_71 0=0 1=2 2=2 5=1 3=0 13=0 14=0 15=0
Convolution            8_75                 1 1 7_71 8_75 0=256 1=3 2=1 3=1 4=1 5=0 6=294912
BatchNorm              8_75_bn              1 1 8_75 8_75_bn 0=256 1=.00001
ReLU                   8_75_bn_leaky        1 1 8_75_bn 8_75_bn_leaky 0=0.1
Split                  8_75_bn_leaky_split  1 2 8_75_bn_leaky 8_75_bn_leaky_split_0 8_75_bn_leaky_split_1
Pooling                9_83                 1 1 8_75_bn_leaky_split_0 9_83 0=0 1=2 2=2 5=1 3=0 13=0 14=0 15=0
Convolution            10_87                1 1 9_83 10_87 0=512 1=3 2=1 3=1 4=1 5=0 6=1179648
BatchNorm              10_87_bn             1 1 10_87 10_87_bn 0=512 1=.00001
ReLU                   10_87_bn_leaky       1 1 10_87_bn 10_87_bn_leaky 0=0.1
Pooling                11_95                1 1 10_87_bn_leaky 11_95 0=0 1=2 2=1 5=1 3=0 13=0 14=0 15=0
Convolution            12_99                1 1 11_95 12_99 0=1024 1=3 2=1 3=1 4=1 5=0 6=4718592
BatchNorm              12_99_bn             1 1 12_99 12_99_bn 0=1024 1=.00001
ReLU                   12_99_bn_leaky       1 1 12_99_bn 12_99_bn_leaky 0=0.1
Convolution            13_109               1 1 12_99_bn_leaky 13_109 0=256 1=1 2=1 3=1 4=0 5=0 6=262144
BatchNorm              13_109_bn            1 1 13_109 13_109_bn 0=256 1=.00001
ReLU                   13_109_bn_leaky      1 1 13_109_bn 13_109_bn_leaky 0=0.1
Split                  13_109_bn_leaky_split 1 2 13_109_bn_leaky 13_109_bn_leaky_split_0 13_109_bn_leaky_split_1
Convolution            14_117               1 1 13_109_bn_leaky_split_0 14_117 0=512 1=3 2=1 3=1 4=1 5=0 6=1179648
BatchNorm              14_117_bn            1 1 14_117 14_117_bn 0=512 1=.00001
ReLU                   14_117_bn_leaky      1 1 14_117_bn 14_117_bn_leaky 0=0.1
Convolution            15_125               1 1 14_117_bn_leaky 15_125 0=18 1=1 2=1 3=1 4=0 5=1 6=9216
Convolution            18_147               1 1 13_109_bn_leaky_split_1 18_147 0=128 1=1 2=1 3=1 4=0 5=0 6=32768
BatchNorm              18_147_bn            1 1 18_147 18_147_bn 0=128 1=.00001
ReLU                   18_147_bn_leaky      1 1 18_147_bn 18_147_bn_leaky 0=0.1
Interp                 19_155               1 1 18_147_bn_leaky 19_155 0=1 1=2.f 2=2.f
Concat                 20_158               2 1 19_155 8_75_bn_leaky_split_1 20_158
Convolution            21_161               1 1 20_158 21_161 0=256 1=3 2=1 3=1 4=1 5=0 6=884736
BatchNorm              21_161_bn            1 1 21_161 21_161_bn 0=256 1=.00001
ReLU                   21_161_bn_leaky      1 1 21_161_bn 21_161_bn_leaky 0=0.1
Convolution            22_169               1 1 21_161_bn_leaky 22_169 0=18 1=1 2=1 3=1 4=0 5=1 6=4608
Yolov3DetectionOutput  detection_out        2 1 15_125 22_169 output 0=1 1=3 2=0.250000 -23304=12,10.000000,14.000000,23.000000,27.000000,37.000000,58.000000,81.000000,82.000000,135.000000,169.000000,344.000000,319.000000 -23305=6,3.000000,4.000000,5.000000,0.000000,1.000000,2.000000 -23306=2,34.666668,16.000000

到这看起来都没什么问题。

step2：ncnnoptimize 报错如下：

super@super-All-Series:~/ncnn/build/tools$ ./ncnnoptimize darknet/yolov3-tiny.param darknet/yolov3-tiny.bin yolov3-tiny-opt-fp32.param yolov3-tiny-opt-fp32.bin 0
fuse_convolution_batchnorm 0_25 0_25_bn
fuse_convolution_batchnorm 2_37 2_37_bn
fuse_convolution_batchnorm 4_49 4_49_bn
fuse_convolution_batchnorm 6_61 6_61_bn
fuse_convolution_batchnorm 8_73 8_73_bn
fuse_convolution_batchnorm 10_85 10_85_bn
fuse_convolution_batchnorm 12_97 12_97_bn
fuse_convolution_batchnorm 13_107 13_107_bn
fuse_convolution_batchnorm 14_115 14_115_bn
fuse_convolution_batchnorm 18_145 18_145_bn
fuse_convolution_batchnorm 21_159 21_159_bn
fuse_convolution_activation 0_25 0_25_bn_leaky
fuse_convolution_activation 2_37 2_37_bn_leaky
fuse_convolution_activation 4_49 4_49_bn_leaky
fuse_convolution_activation 6_61 6_61_bn_leaky
fuse_convolution_activation 8_73 8_73_bn_leaky
fuse_convolution_activation 10_85 10_85_bn_leaky
fuse_convolution_activation 12_97 12_97_bn_leaky
fuse_convolution_activation 13_107 13_107_bn_leaky
fuse_convolution_activation 14_115 14_115_bn_leaky
fuse_convolution_activation 18_145 18_145_bn_leaky
fuse_convolution_activation 21_159 21_159_bn_leaky
shape_inference
总线错误 (核心已转储)

感谢解答~

infrontofme commented 4 years ago

又测试了一下yolov4-tiny模型没有报错

eheroqiu commented 4 years ago

同yolo3-tiny , 相同的问题bus error(core dumped)~

agrichron commented 3 years ago

转换后的param和原始的cfg参数有出入。

vicwer commented 3 years ago

相同问题，yolov3-tiny可以，yolov4-tiny不可以，有解决吗？

Tencent / ncnn

ncnnoptimize yolov3-tiny error #2072