Refine this repo #11
Refine this repo #11
Conversation
|
|
||
| ## Model compression |
There was a problem hiding this comment.
I don't understand why separate the rounding and pruning
There was a problem hiding this comment.
how about set Deployment optimization methods to Tools and move the rounding to Model compression
There was a problem hiding this comment.
As with the merge batch normalization, the rounding method is used to process the model before the inference. So I put it in the tool directory.
benchmark/README.md
Outdated
|
|
||
| | Model | Xiaomi MI 5(ST) | | ||
| |-----------| ----------- | | ||
| | Mobilenet | 122.607(ms) | |
There was a problem hiding this comment.
need to specify the type of mobilenet (1.0 or 0.75 , 0.5 0.25) and input size(224* 224 * 3)
benchmark/model/mobilenet.py
Outdated
| name="output", input=net, size=classdim, act=paddle.activation.Softmax()) | ||
|
|
||
| Inputs("image") | ||
| Outputs("output") |
There was a problem hiding this comment.
This config can not be used in v2 api, all mobilenet config in this repo should be the same specifications, and could use for v1 and v2 api, here is a example https://gist.github.com/NHZlX/204da5e89a1263ab6953a669401e37d5 , and this need to be discussed.
There was a problem hiding this comment.
I temporarily remove this config file.
No description provided.