Geonuk Kim

Training from scratch, w/o using Pretrained DeiT?

3

Thanks for sharing the source codes! I found that you exploited **'Pretrained weight file of DeiT'** instead of training from scratch. However, i see you emphasize 'Efficiency' of your model....

Thanks for the source code. I want to use coco2017 instead of coco2014. I knew that the difference between coco2017 and coco2014 is data split (2014minival == 2017test). Is there...

Thanks for nice paper and releasing source code. Unfortunately, i got confused with your description about install. I see that you modified original detectron2 package. (i.e. dataset define) Thus, there...

Paper’s Result is not reproducible

9

As many issues in here said the results on the paper are not reproducible. The performance shows high variance along the randomly sampled k-shot images at 2nd training phase. I...

I found un-fair data usage

4

Hi there. I well read uploaded source code. I found you set k-shot masked sample with k-'image' not 'instance'. It might be okay that there is binary mask for input...

error during test on phase 1

1

Thx to your released code. Unfortunately i countered an error test on phase 1 (for base class) line 332, in inds = torch.nonzero(scores[:, j] > thresh).view(-1) IndexError: index 16 is...

How can i select a certain word vector?

1

Hi, i'm newbie for NLP but i'd like to select some category's word vector. I want word vectors of MS-COCO class's name, such as 'Person', 'Bus', 'Bird'... I downloaded pretrained...

Novel samples for finetuning

1

Hello, thanks for you nice paper and released source code. I wonder how did you select K-shot samples in Pascal VOC dataset for novel classes. Is it same with FSRW(ICCV...

Nice paper and thanks for the source code! I noticed that you use 'class dropout' which is not mentioned in your paper. What's the purpose of the dropout function with...

Hello, nice work and thanks for the releasing source code! As there was same issue before.. i am confused that **which config file I should choose to obtain the** [pretrained...