YOLO v2 Net
data · Data
transform param
crop size: 416
mirror: true
data param
source: PASCAL VOC 2007 train + 2012 train & test
batch size: 32
blob shapes
data: [ 1, 3, 416, 416 ]
label: [ 1 ]
activation: leakyrelu
transform param
crop size: 416
mirror: true
data param
source: PASCAL VOC 2007 train + 2012 train & test
batch size: 32
blob shapes
data: [ 1, 3, 416, 416 ]
label: [ 1 ]
activation: leakyrelu
Input image size: 416x416
Input image amount: 3
network model: Convolution
layer: 0
Output amount: 32
Stride: 1
weights size: 3x3
Output image size: 416x416
Input image amount: 3
network model: Convolution
layer: 0
Output amount: 32
Stride: 1
weights size: 3x3
Output image size: 416x416
Input image size: 416x416
Input image amount: 32
network model: MaxPool
layer: 1
Output amount: 32
Stride: 2
weights size: 2x2
Output image size: 208x208
Input image amount: 32
network model: MaxPool
layer: 1
Output amount: 32
Stride: 2
weights size: 2x2
Output image size: 208x208
Input image size: 208x208
Input image amount: 32
network model: Convolution
layer: 2
Output amount: 64
Stride: 1
weights size: 3x3
Output image size: 208x208
Input image amount: 32
network model: Convolution
layer: 2
Output amount: 64
Stride: 1
weights size: 3x3
Output image size: 208x208
Input image size: 208x208
Input image amount: 64
network model: MaxPool
layer: 3
Output amount: 64
Stride: 2
weights size: 2x2
Output image size: 104x104
Input image amount: 64
network model: MaxPool
layer: 3
Output amount: 64
Stride: 2
weights size: 2x2
Output image size: 104x104
Input image size: 104x104
Input image amount: 64
network model: Convolution
layer: 4
Output amount: 128
Stride: 1
weights size: 3x3
Output image size: 104x104
Input image amount: 64
network model: Convolution
layer: 4
Output amount: 128
Stride: 1
weights size: 3x3
Output image size: 104x104
Input image size: 104x104
Input image amount: 64
network model: Convolution
layer: 5
Output amount: 64
Stride: 1
weights size: 3x3
Output image size: 104x104
Input image amount: 64
network model: Convolution
layer: 5
Output amount: 64
Stride: 1
weights size: 3x3
Output image size: 104x104
Input image size: 104x104
Input image amount: 64
network model: Convolution
layer: 6
Output amount: 128
Stride: 1
weights size: 3x3
Output image size: 104x104
Input image amount: 64
network model: Convolution
layer: 6
Output amount: 128
Stride: 1
weights size: 3x3
Output image size: 104x104
Input image size: 104x104
Input image amount: 128
network model: MaxPool
layer: 7
Output amount: 128
Stride: 2
weights size: 2x2
Output image size: 52x52
Input image amount: 128
network model: MaxPool
layer: 7
Output amount: 128
Stride: 2
weights size: 2x2
Output image size: 52x52
Input image size: 52x52
Input image amount: 128
network model: Convolution
layer: 8
Output amount: 256
Stride: 1
weights size: 3x3
Output image size: 52x52
Input image amount: 128
network model: Convolution
layer: 8
Output amount: 256
Stride: 1
weights size: 3x3
Output image size: 52x52
Input image size: 52x52
Input image amount: 256
network model: Convolution
layer: 9
Output amount: 128
Stride: 1
weights size: 3x3
Output image size: 52x52
Input image amount: 256
network model: Convolution
layer: 9
Output amount: 128
Stride: 1
weights size: 3x3
Output image size: 52x52
Input image size: 52x52
Input image amount: 128
network model: Convolution
layer: 10
Output amount: 256
Stride: 1
weights size: 3x3
Output image size: 52x52
Input image amount: 128
network model: Convolution
layer: 10
Output amount: 256
Stride: 1
weights size: 3x3
Output image size: 52x52
Input image size: 52x52
Input image amount: 256
network model: MaxPool
layer: 11
Output amount: 256
Stride: 2
weights size: 2x2
Output image size: 26x26
Input image amount: 256
network model: MaxPool
layer: 11
Output amount: 256
Stride: 2
weights size: 2x2
Output image size: 26x26
Input image size: 26x26
Input image amount: 256
network model: Convolution
layer: 12
Output amount: 512
Stride: 1
weights size: 3x3
Output image size: 26x26
Input image amount: 256
network model: Convolution
layer: 12
Output amount: 512
Stride: 1
weights size: 3x3
Output image size: 26x26
Input image size: 26x26
Input image amount: 512
network model: Convolution
layer: 13
Output amount: 256
Stride: 1
weights size: 3x3
Output image size: 26x26
Input image amount: 512
network model: Convolution
layer: 13
Output amount: 256
Stride: 1
weights size: 3x3
Output image size: 26x26
Input image size: 26x26
Input image amount: 256
network model: Convolution
layer: 14
Output amount: 512
Stride: 1
weights size: 3x3
Output image size: 26x26
Input image amount: 256
network model: Convolution
layer: 14
Output amount: 512
Stride: 1
weights size: 3x3
Output image size: 26x26
Input image size: 26x26
Input image amount: 512
network model: Convolution
layer: 15
Output amount: 256
Stride: 1
weights size: 3x3
Output image size: 26x26
Input image amount: 512
network model: Convolution
layer: 15
Output amount: 256
Stride: 1
weights size: 3x3
Output image size: 26x26
Input image size: 26x26
Input image amount: 256
network model: Convolution
layer: 16
Output amount: 512
Stride: 1
weights size: 3x3
Output image size: 26x26
Input image amount: 256
network model: Convolution
layer: 16
Output amount: 512
Stride: 1
weights size: 3x3
Output image size: 26x26
Input image size: 26x26
Input image amount: 512
network model: MaxPool
layer: 17
Output amount: 512
Stride: 2
weights size: 2x2
Output image size: 13x13
Input image amount: 512
network model: MaxPool
layer: 17
Output amount: 512
Stride: 2
weights size: 2x2
Output image size: 13x13
Input image size: 13x13
Input image amount: 512
network model: Convolution
layer: 18
Output amount: 1024
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image amount: 512
network model: Convolution
layer: 18
Output amount: 1024
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image size: 13x13
Input image amount: 1024
network model: Convolution
layer: 19
Output amount: 512
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image amount: 1024
network model: Convolution
layer: 19
Output amount: 512
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image size: 13x13
Input image amount: 512
network model: Convolution
layer: 20
Output amount: 1024
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image amount: 512
network model: Convolution
layer: 20
Output amount: 1024
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image size: 13x13
Input image amount: 1024
network model: Convolution
layer: 21
Output amount: 512
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image amount: 1024
network model: Convolution
layer: 21
Output amount: 512
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image size: 13x13
Input image amount: 512
network model: Convolution
layer: 22
Output amount: 1024
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image amount: 512
network model: Convolution
layer: 22
Output amount: 1024
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image size: 13x13
Input image amount: 1024
network model: Convolution
layer: 23
Output amount: 1024
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image amount: 1024
network model: Convolution
layer: 23
Output amount: 1024
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image size: 13x13
Input image amount: 1024
network model: Convolution
layer: 24
Output amount: 1024
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image amount: 1024
network model: Convolution
layer: 24
Output amount: 1024
Stride: 1
weights size: 3x3
Output image size: 13x13
layer: 25
the 16th layer output
the 16th layer output
Input image size: 26x26
Input image amount: 512
network model: Convolution
layer: 26
Output amount: 64
weights size: 3x3
Output image size: 26x26
Input image amount: 512
network model: Convolution
layer: 26
Output amount: 64
weights size: 3x3
Output image size: 26x26
Input image size: 26x26
Input image amount: 64
network model: Reorganization
layer: 27
Output amount: 256
Output image size: 13x13
Input image amount: 64
network model: Reorganization
layer: 27
Output amount: 256
Output image size: 13x13
network model: concat 合併
合併layer: 27 24
layer: 28
合併layer: 27 24
layer: 28
Input image size: 13x13
Input image amount: 1280
network model: Convolution
layer: 29
Output amount: 1024
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image amount: 1280
network model: Convolution
layer: 29
Output amount: 1024
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image size: 13x13
Input image amount: 1024
network model: Convolution
layer: 30
Output amount: 1024
Stride: 1
weights size: 3x3
Output image size: 13x13
Input image amount: 1024
network model: Convolution
layer: 30
Output amount: 1024
Stride: 1
weights size: 3x3
Output image size: 13x13
Region ·Region
Dectection
| Layer | Type | Filters | Size/Stride | Output size |
|---|---|---|---|---|
| 0 | Convolutional | 32 | 3x3/1 | 416x416 |
| 1 | Maxpool | 2x2/2 | 208x208 | |
| 2 | Convolutional | 64 | 3x3/1 | 208x208 |
| 3 | Maxpool | 2x2/2 | 104x104 | |
| 4 | Convolutional | 128 | 3x3/1 | 104x104 |
| 5 | Convolutional | 64 | 3x3/1 | 104x104 |
| 6 | Convolutional | 128 | 3x3/1 | 104x104 |
| 7 | Maxpool | 2x2/2 | 52x52 | |
| 8 | Convolutional | 256 | 3x3/1 | 52x52 |
| 9 | Convolutional | 128 | 1x1/1 | 52x52 |
| 10 | Convolutional | 256 | 3x3/1 | 52x52 |
| 11 | Maxpool | 2x2/2 | 26x26 | |
| 12 | Convolutional | 512 | 3x3/1 | 26x26 |
| 13 | Convolutional | 256 | 3x3/1 | 26x26 |
| 14 | Convolutional | 512 | 3x3/1 | 26x26 |
| 15 | Convolutional | 256 | 1x1/1 | 26x26 |
| 16 | Convolutional | 512 | 3x3/1 | 26x26 |
| 17 | Maxpool | 2x2/2 | 13x13 | |
| 18 | Convolutional | 1024 | 3x3/1 | 13x13 |
| 19 | Convolutional | 512 | 1x1/1 | 13x13 |
| 20 | Convolutional | 1024 | 3x3/1 | 13x13 |
| 21 | Convolutional | 512 | 1x1/1 | 13x13 |
| 22 | Convolutional | 1024 | 3x3/1 | 13x13 |
| 23 | Convolutional | 1024 | 3x3/1 | 13x13 |
| 24 | Convolutional | 1024 | 3x3/1 | 13x13 |
| 25 | Route | 16 | ||
| 26 | Convolutional | 64 | 1x1/1 | 26x26 |
| 27 | Reorg | /2 | 13x13 | |
| 28 | Route | 27 24 | ||
| 29 | Convolutional | 1024 | 3x3/1 | 13x13 |
| 30 | Convolutional | 425 | 1x1/1 | 13x13 |
| 31 | Detection |
Original
Output
