issues
search
ChenRocks
/
UNITER
Research code for ECCV 2020 paper "UNITER: UNiversal Image-TExt Representation Learning"
https://arxiv.org/abs/1909.11740
777
stars
109
forks
source link
About fp16 and location features.
#93
Closed
EstherBear
closed
2 years ago
EstherBear
commented
2 years ago
Hi, I have a problem in training.
The last elements of location features (w x h in [x1, y1, x2, y2, w, h, w × h]) will cause overflow in fp16 training mode. So how do you train in fp16?
And what is the meaning of "normalized" in the location feature? Could you provide an example?
Thanks!
Hi, I have a problem in training.
Thanks!