Hi, I want to ask that in the paper, you mentioned that:
index 0 corresponds to the probability of dropping token
index 1 corresponds to the probability of keeping token
However, when I trace the code:
|
score = pred_score[:,:,0] |
this line seems to indicate that index 0 is the score of keeping token, so I'm confused about the actual implementation in your work, Thanks !!!
Hi, I want to ask that in the paper, you mentioned that:
index 0 corresponds to the probability of dropping token
index 1 corresponds to the probability of keeping token
However, when I trace the code:
DynamicViT/models/dyvit.py
Line 444 in 1322e62
this line seems to indicate that index 0 is the score of keeping token, so I'm confused about the actual implementation in your work, Thanks !!!