Marco Ciccone comments

Results 20 comments of


                                            Marco Ciccone

Use Beholder plugin in TensorboardX?

Is there anyone working on the porting? I have a pretty much working version, but it's definitely too slow. If you want I can open a PR and discuss on...

Use Beholder plugin in TensorboardX?

Hi @lanpa, yes I've tried the demo, but it was missing al lot of features such as resizing of different shapes to be displayed all together. I think I implemented...

Conversion of inception model from tensorflow to mxnet

Can you please explain the conversion process? Thanks

Saved model does not work in onlyTest mode but it works in Training mode

I think that the saving model procedure doesn't correctly save all the parameters. The error might occur when you train a BatchNormalized model where you have to save also a...

Support for haiku models with non-trainable state

Thanks @stheertha! I agree that tracking statistics is nontrivial in FL. For the moment I've overcome the issue by replacing batchnorm with groupnorm and it seems to be working fine....

Support for gldv2 and inaturalist datasets

Hi @stheertha and thanks for the support! I'll have a look at the SQLite stuff and get back to you. I went a different way for the moment (just to...

Support for gldv2 and inaturalist datasets

Hi @kho! By looking at images in the tfrecord of a randomly sampled client, I see that images have different height/width. Images should be then randomly cropped to 299x299 (inaturalist)...

Support for gldv2 and inaturalist datasets

Thanks for your answer! I think option 1 is the way to go to ensure enough data variability (multiple crops for each image). However, gldv2 image sizes span from 300...

index out of range in THTensorMath.c

I think that this is a memory error due to the huge size of the embedding lookup table or the one hot vector that you're using. If you use a...

has_aux option in custom_root

Tagging @froystig in case he has a solution involving `jax.eval_shape`