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

Weird glosses in the annotation of phoenix dataset #13

Closed sunke123 closed 2 years ago

sunke123 commented 2 years ago

Hi @ycmin95 , recently, I checked the annotation of phoenix dataset and the gloss dictionary generated during the progress of data preparation. There are many weird glosses, such as "ON", "OFF", "LEFTHAND" ... image I wonder whether we should keep these weird glosses in the label... Any advice?

ycmin95 commented 2 years ago

Hi, if you look the evaluation script, you will find that most stop words and synonyms are filtered or merged, which can be found here. And this script is modified from the original evaluation script provided in PHOENIX14 (./phoenix-2014-multisigner/evaluation/evaluatePhoenix2014.sh). The left "weired glosses" are ON and OFF, from my understanding, they means pause of sentence, which is also useful for understanding. And the recognition results on these two glosses are acceptable, so I keep the same evaluation method as the dataset provided.

sunke123 commented 2 years ago

Got it~ Thanks~