search

VIPL-SLP / VAC_CSLR

Visual Alignment Constraint for Continuous Sign Language Recognition. ( ICCV 2021)
https://openaccess.thecvf.com/content/ICCV2021/html/Min_Visual_Alignment_Constraint_for_Continuous_Sign_Language_Recognition_ICCV_2021_paper.html
Apache License 2.0
116 stars 19 forks source link

About the resnet18 backbone #2

Closed chrismcruiz closed 2 years ago

chrismcruiz commented 2 years ago

Hi, I was wondering if the resnet18 backbone that you guys posted here is from the non-iterative or the iterative approach. I suppose is from the non-iterative but I'm not quite sure.

Finally, can I use the feature extractor function that you provided into another dataset? and if so, what should be the set up?

Thanks in advance.

ycmin95 commented 2 years ago

@chrismcruiz Hi, the released model is based on the proposed VAC, which is the non-iterative approach. We have not conducted cross dataset experiments yet. From my understanding, the feature extractor can be applied on another dataset, you need to change the dataloader and adopt the "features" mode, which will extract frame-wise features. However, the posture distribution from different datasets may be different, maybe it will not achieve optimal performance, it is better to finetune or train on another dataset.