Ting Chen
Ting Chen
Yes, you'll still need that to create two augmented views. On Tue, Jun 29, 2021 at 4:25 AM ines321 ***@***.***> wrote: > In the official paper of SimCLR, the authors...
> as the training loss declines, the entropy of the distribution increases rather than decreases. This seems plausible because near convergence, the probability scores for all the negative examples are...
It's the prediction accuracy of positive examples among all candidates within the mini-batch. On Fri, Jan 28, 2022 at 6:51 PM slala2121 ***@***.***> wrote: > Could you explain how the...
se_ratio is always 0 so you can ignore. in the released simclrv2 models, sk_ratio is set either 0 (wo sk) or 0.0625 (w sk).
Not entirely sure, but it seems `model_dir` is not compatible with your file system. Maybe try running locally (e.g. unset `tpu_name`) and see if there's still issue with the file...
Did you set `use_tpu=False` when you disable `tpu_name`?
@saxenasaurabh, who created TF2 and converted SavedModels from TF1, probably knows more about this.
Yes, you're right, this should be a typo. It shouldn't affect default behavior since we set both height and width to image_size, but it would be a bug if width...
`/tmp/simclr_test` is the folder where the model checkpoints will be saved, and it is a temporary folder for unix based system. But you could set it to any other writable...
we use random crop / flip during fine-tuning, but didn't find color augmentation very useful in fine-tuning.