-
Resnet Dorefa activation quantization incorrect in the model
-
我升级TF到0.12之后,程序可以初始化可以完成,但是训练进度一直为 0%, 请问一下是什么原因呢? 谢谢了
![qq 20161216145437](https://cloud.githubusercontent.com/assets/8264748/21254344/a6d187b4-c39f-11e6-841f-47dd7e09a2c8.png)
-
### Describe the bug
expression "n = 2 ** self.precision - 1" seems wrong
it should be "n = 2 ** self.precision"
or "n = 2 ** (self.precision - 1)"
### To Reproduce
### Expected behavior
…
-
### Feature motivation
Current DoReFa quantizer is limited to [0,1] activations, and actually seems to have a bug in the quantization formula
### Feature description
New quantizer "ClipAndQua…
-
@SpursLipu ,
您好!感谢您的分享,受益匪浅!请问,您有做过yolov3的8位量化实验么?谢谢!
祝好!
-
Thank you for your work!
I have tried to quantize ABCNet but have been unsuccessful due to the below reasons:
- Some code not supported by torch.jit
- Some errors by Quant/DeQuant
Could you p…
-
If I want to train a dorefa-Net quantization on alexnet , is the command line like this ?
`python compress_classifier.py -a alexnet /ImageNet_Share/ --compress=/distiller/examples/quantization/quant_…
-
I too try to replicate the experiment in:
https://github.com/czhu95/ternarynet/blob/master/README.md
Using a very old release of TF/tensorpack:
+Ubuntu 16.04
+Python 2.7
+TF release 1.1.0
+ …
-
https://github.com/tensorpack/tensorpack/blob/b28cfa8a50b732651b3711d09afd0818083861b2/examples/DoReFa-Net/resnet-dorefa.py#L107
What will happen if without this line?
-
The quantization aware training with PACT quantizer is taking an unreasonably long time compared to training with full precision or with other quantizers. Any clue on why is this happening? In PACT, t…