HyperGAN
Search…
resizable_stack
1
description: 'layer resizable_stack for configurable component'
2
---
3
​
4
# resizable_stack layer
5
​
6
`resizable_stack` allows for variable size outputs on the generator. A conv stack is repeated until the output size is reached.
7
​
8
If you specify "segment_softmax" this repeats the pattern:
9
upsample
10
normalize(expects style vector named 'w')
11
conv ...
12
activation(before last layer)
13
​
14
and ends in:
15
segment_softmax output_channels
16
​
17
## arguments
18
* layer type. Defaults to "segment_softmax"
19
## optional arguments
20
* segment_channels - The number of channels before segment_softmax. Defaults to 5
21
* max_channels - The most channels for any conv. Default 256
22
* style - the style vector to use. Default "w"
23
## input size
24
​
25
Any 4-d tensor
26
​
27
## output size
28
​
29
[B, output channels, output height, output width]
30
​
31
## syntax
32
​
33
```json
34
"resizable_stack segment_softmax"
35
```
36
​
37
## examples
38
​
39
```json
40
"identity name=w",
41
"linear 8*8*256",
42
"relu",
43
"resizable_stack segment_softmax",
44
"hardtanh"
45
```
Copied!
Last modified 1yr ago
Copy link