CAI, Rizhao

The > My environment is RTX 2080, ubuntu 18.04, python 3.7, cuda 10.0 The problems maybe because of python3.7. python3.5 may help. Btw, do you manage to solve the problem...

`_C = CN(new_allowed=True)` It works for me.

Same question

> Hi @gitti123, > > > I think it is because of the expectation that the batchnormalization layer will come after the convolutional layer and not the other way round...

> How to save and load models？ SAME QUESTION. How to save a model and load it? Have you found the way?

> How to save and load models？ class SKlearnBaseClassifier(BaseClassifierWrapper): def _load_model_from_disk(self, cache_path): return joblib.load(cache_path) def _save_model_to_disk(self, clf, cache_path): joblib.dump(clf, cache_path) you can refer to this: 

I also wonder that, at [line](https://github.com/HobbitLong/SupContrast/blob/331aab5921c1def3395918c05b214320bef54815/losses.py#L89 )  Isn't that the distance between positive pairs are being the denominator because of `exp_logits.sum(1, keepdim=True)`?