ByungSoo Ko comments

Results 8 comments of


                                            ByungSoo Ko

Process hangs at the end

The stack trace is as below. I guess it was stuck by below error. ```bash Traceback (most recent call last): File "/root/anaconda3/envs/crawl/lib/python3.8/multiprocessing/pool.py", line 851, in next item = self._items.popleft() IndexError:...

Process hangs at the end

@rom1504 Thanks! I will test it and let you know.

Process hangs at the end

@rom1504 So, I tested it with 1.20.2 version and it still hangs at the end. Below is the output when I terminate the process by force. ``` worker - success:...

What is the meaning of 'non-conventional usage' of backbone at Table 7 in the Paper?

Thanks for the interest in our paper. For your question, it means 'discarding the down-sampling operation between stage3 and stage4' in section 3.1 of the paper. So, you are right...

Some questions about the paper

Sorry for late reply. 1. It is actually a well-known fact that you could get lower performance with FC layer than GD. It is because FC layer is for dimensional...

"adamw" optimizer + weight decay = poor generations

yeap, let's keep it open since it's from the paper :)

"adamw" optimizer + weight decay = poor generations

@robvanvolt the default weight_decay is 0, but DALLE paper used 4.5*10-2.

can not reproduce your results. trained from your released pre-trained vit-base model

I had the same problem when I finetuned the ViT-B/32 with 1 node or 4 nodes. But when I finetuned with 2 nodes, I got 83.784 top-1 acc. Try finetuning...