Updated inceptionV3 to accept different sized images #718
Conversation
The update allows inceptionV3 to process images larger or smaller than prescribed image size (299x299).
Codecov Report
@@ Coverage Diff @@
## master #718 +/- ##
==========================================
+ Coverage 38.99% 39.61% +0.61%
==========================================
Files 29 29
Lines 2721 2724 +3
Branches 411 430 +19
==========================================
+ Hits 1061 1079 +18
+ Misses 1579 1571 -8
+ Partials 81 74 -7
Continue to review full report at Codecov.
|
fmassa
left a comment
fmassa
left a comment
There was a problem hiding this comment.
Thanks for the PR!
I have a few comments, but I'd be willing to merge the PR if you address those.
| # 8 x 8 x 2048 | ||
| x = F.avg_pool2d(x, kernel_size=8) | ||
| adaptiveAvgPoolWidth = x.shape[2] | ||
| x = F.avg_pool2d(x, kernel_size=adaptiveAvgPoolWidth) |
There was a problem hiding this comment.
isn't it simpler to just use F.adaptive_avg_pool2d(x, 1) here?
There was a problem hiding this comment.
Adaptive avg pool wasn't working in my local implementation due to pytorch version incompatibility, as you would be knowing from the issue (#696), so I implemented a workaround around it. From your above question it seems that that the latest pytorch release has the Adaptive avg pool so your suggestion is indeed a better implementation. Should I make this change in my PR?
There was a problem hiding this comment.
Please have a look at the new PR (#744 (comment)).
| # 5 x 5 x 128 | ||
| x = self.conv1(x) | ||
| adaptiveAvgPoolWidth = x.shape[2] | ||
| x = F.avg_pool2d(x, kernel_size=adaptiveAvgPoolWidth) |
There was a problem hiding this comment.
Same here, isn't it simpler to just use the adaptive version?
|
superseeded by #744 |
3 participants
The update allows inceptionV3 to process images larger or smaller than prescribed image size (299x299). Will be useful while finetuning or testing on different resolution images.