Wouter Van Gansbeke

Hi @ZhiyuanDang, I noticed that quite a lot of people have troubles with the implementation of KMeans. I might upload a script after the NeurIPS deadline. For now, did you...

Hi @ZhiyuanDang, You should obtain more or less the same accuracy for the train and test set on ImageNet (distributions are very similar). I always report the numbers on the...

Hi @ZhiyuanDang, FYI, I double checked and get the same results as in the paper.

Make sure that you L2 normalize the features and that you report the results on the test set (use the train set to fit the kmeans). It doesn't even matter...

Hi @bwbelljr @eyalho, Your contribution is very much appreciated, @bwbelljr! I haven't found the time to check this pull request yet, but it looks useful. I want to make sure...

Not sure I understand your question. The ground truth (i.e. "dense depth") can not be used as an input during training. We need to predict it after all. What would...

Hi @q5390498 , Ok makes sense. I believe that would work fine. Maybe quickly try it out. However, keep in mind that fusion takes place at multiple stages in the...

Since I don't know any details about your dataset (e.g. imbalance), I can't help much. We only provide results for CIFAR, STL and ImageNet. Try experimenting with these datasets first...

You might want to reduce the entropy weight in this case. Set it to 1 or 2. Note that we applied stronger augmentations for the SCAN part (strong color augmentations...

Yes, I meant `data/augment.py`. That should be the only place.