joe-siyuan-qiao / WeightStandardization

Standardizing weights to accelerate micro-batch training

joe-siyuan-qiao/WeightStandardization Issues

Implementation question in Tensorflow
Updated a year ago4
Network's performance decreases after adopting WS
Updated 3 years ago5
WS with ConvTranspose2d?
Updated 3 years ago
changing the position of epsilon
Updated 3 years ago2
It just explodes!!!
Updated 3 years ago1
How to apply WS in video understanding?
Closed 3 years ago
Can i apply the same method to 3D conv?
Updated 4 years ago5
Possible typo in the paper and some other questions
Updated 4 years ago
consistent mean calculation with std calculation
Updated 4 years ago2
Questions about the definition of WS and a possible typo in the paper.
Closed 5 years ago7
How to load weights of Weight Standardization model?
Updated 4 years ago
Error with multi_gpu_model on Keras
Updated 4 years ago
WD or WD+GN with fp16
Updated 4 years ago3
It seems that same method has been proposed before
Updated 4 years ago2
AttributeError: 'module' object has no attribute 'reduce_mean'
Closed 5 years ago4
Possible Keras Implementation doesn't seem to work
Closed 5 years ago7
tf.math.reduce_mean(kernel, axis=[0, 1, 2],
Closed 5 years ago2
Using BN+WS in ResNet50 similar or slightly worse than using BN only in ResNet50 for ImageNet classification
Closed 5 years ago4
the loss is nan
Updated 5 years ago6
It seems to have no effect on this work.
Updated 5 years ago1
WS of Deformable Convolution.
Updated 5 years ago3
batch size and iteration
Updated 5 years ago2
About pretrained model R101-GN-WS
Closed 5 years ago4
In my model, GN+WS is worse
Closed 5 years ago4
about train and inference
Updated 5 years ago1
Quesions on accuracy gain of WS?
Closed 5 years ago2
In my test,gn+ws result is worse
Closed 5 years ago2
Conv1d Version?
Closed 5 years ago5