yates sun

Questions about cityscapes validation

1

You use the resolution of 769 * 769 when training cityscapes but use 1025 * 2049 when validation。 If there be a gap between the two results

Distribute in Imagenet

3

When I use Distribution in imagenet with one Node, there is three print at the same time. Just like this.  Is it right?

I feel uncertain about the candidate operations in the reduce layers.

2

Would you tell me that is there a skip operation in the reduce layers? Thank you .

Replace AffinityNet with IRN

2

Dear YudeWang, Thanks for your code! Do you have replaced AffinityNet with IRN before? I get a worse result when I replaced AffinityNet with IRN. Could you give me some...

about fp16

5

When I use fp16 (16-bit float) and multi-gpu training,the code will wait in SyncBN(comm.py). 

The QuantizersSwitcher has a bug, where in the following situation, calling enable_quantizers() only once does not achieve the expected goal; it must be called twice: - When the quantizers are...