@MattiaSegu@wightmanr Could you clarify what type of anti-aliasing is being discussed here? I've looked through the MobileNet-V4 papers but haven't found explicit mentions of the anti-aliasing approach being used. Is this a specific downsampling technique or something else?
@almenal99@treyhunner It's a bit silly but I some times do Path.str = Path.__str__ after I import pathlib. That way, I can write Path.str() which is easier than to move back and forth to wrap it.
@elie@TensorFlow@fchollet@JeffDean Self-supervised pre-training is very exciting, but doesn't this blog post underestimate the test set accuracy you can expect when training cifar10 from scratch by quite a bit?
We thought we had defeated the natural world because we defeated the animals and plants, but it turned out that they were only our fellow-travelers, struggling to survive. The real natural world is dust and fire and flood and hard radiation. It never sleeps and it always hungers.