Add TIN model #53
Add TIN model #53
Conversation
Codecov Report
@@ Coverage Diff @@
## master #53 +/- ##
==========================================
- Coverage 86.50% 84.41% -2.09%
==========================================
Files 72 73 +1
Lines 4030 4178 +148
Branches 623 633 +10
==========================================
+ Hits 3486 3527 +41
- Misses 440 547 +107
Partials 104 104
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
dreamerlin
force-pushed
the
tin
branch
3 times, most recently
from
5f617fd
to
a1cca11
Compare
dreamerlin
force-pushed
the
tin
branch
4 times, most recently
from
295344b
to
5f0d7e3
Compare
| """Initiate the parameters either from existing checkpoint or from | ||
| scratch.""" | ||
| # we set the initial bias of the convolution | ||
| # layer to 0, and the final initial output will be 1.0 |
There was a problem hiding this comment.
what is "final initial" output?
There was a problem hiding this comment.
This comment is the same with the original repo and the paper. It means the initialized conv layer will output 1.0
There was a problem hiding this comment.
what's the initial weights?
There was a problem hiding this comment.
The conv weights are initialized by default using pytorch kaiming_uniform_ or loading from ckpt, the codes only changes the bias as the paper said.
No description provided.