SimCLR

SimCLR copied to clipboard

Added support for TFIW dataset

1

mishragauravgm

exclude_from_weight_decay added

6

sadimanna

Hi, thank you very much for your code. But I have a doubt about the projection head. I think there should be two BN layers i.e. linear layer , BN...

ToaTao

I want to use SSL to my project, [LFW](http://vis-www.cs.umass.edu/lfw/) is only a demo. Model structure, loss function, optimizer is keeping the same as this repo, only one GPU, load ResNet50...

muyuuuu

The loss doesn't decrease when using multi nodes.

10

When i use one node, the code runs well. However, when I use 2 nodes and set the batch_size to 64, the loss is always around 5.545 and doesn't decrease....

lailvlong

Question about multi-gpu training

2

Thanks for sharing such an excellent work： Now I want to use eight GPUs for DDP training on a server. How should I use the script? I am a little...

Allfu

Code explanation in gather.py

1

Hi, Janne The GatherLayer module in gather.py is smart and efficient. I really appreciate this module. I almost understand most of the functions of the codes but still confuse at...

littleredxh

Hello, thanks for your pretty implementation. I think I may find a small bug in your LARS implementation.`trust_ratio = tf.where( tf.greater(w_norm, 0), tf.where( tf.greater(g_norm, 0), (self.eeta * w_norm / g_norm),...

Big-Brother-Pikachu

CIFAR10 accuracy

3

Hi Spijkervet,@Spijkervet have you tried more epochs for your experiment settings, like 500 epochs, the result from the paper has shown that even 256 batchsize could achieve about 93% accuracy...

MrChenFeng

The default encoding of some non-English speaking countries is not utf-8, so the specified encoding must be specified, otherwise decoding errors will occur.

ghost