Changmao Cheng
Changmao Cheng
@D-X-Y @prlz77 I'm faced with the same problem when reproducing the performance of DenseNet-40 on CIFAR100. With the exactly same configuration, the acc of PyTorch version is often 1 point...
@prlz77 I set `cudnn.benchmark = True` and `cudnn.deterministic = True`. Is that ok?
@prlz77 No improvements from my experiments. Thank you anyway.
@boluoweifenda I haven't train it via tensorflow. There are a lot of ways to improve performance if you don't care about the fair comparison, like using dropout, a better lr...
@louyanyang Sorry, I still don't understand the meaning of `num_deformable_group`...In which case `num_deformable_group` > 1 is needed?
@sinAshish Please set `strict=False` when loading state dict as I didn't include parameters of aux_features in the provided checkpoints.
@gpleiss This re-implementation (https://github.com/wandering007/efficient-densenet-pytorch) has good support for `nn.DataParallel`, which may be helpful.
@gpleiss Yes, it works fine. However, there is one thing that I've noticed before and have to mention, though it is out of the scope of this issue. With checkpointing...
@gpleiss It may be tough for now...To be frank, I am still in favor of the previous implementation (v0.3.1) via `_EfficientDensenetBottleneck` class and `_DummyBackwardHookFn` function without touching any C code....
@hjy1312 Thanks! I'll try it.