-
Notifications
You must be signed in to change notification settings - Fork 3.8k
[TOPI] Upsampling op support #772
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
|
@tqchen Is it possible to install scikit-image on the CI server? I need this for upsampling test. scipy also has imresize function, but they recommend skimage.transform.resize, which I use. |
|
I tried adding scikit-image. Unfortunately there are some issue in upgrading docker to add this(mainly related to something about sphinx and python build) So we cannot do this for now until next major upgrade of our docker test image. Can you do a numpy version or scipy one for now? |
|
@tqchen I removed skimage dependency, but I don't understand why CI is still failing? |
|
was due to stale cache in the build, i am cleaning up things and trigger a build again |
topi/python/topi/generic/nn.py
Outdated
|
|
||
|
|
||
| @tvm.target.generic_func | ||
| def schedule_upsampling(outs): |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
since it uses the default injective schedule, we should simply remove schedule_upsampling function and call schedule_injective instead.
tqchen
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
please remove schedule_upsampling as it should be simply schedule_injective
|
I removed schedule_upsampling for now, but if we need bilinear upsampling later, we will need a schedule for upsampling. |
* add upsampling cpu op * add upsampling gpu schedule * add doc for upsampling op add more doc * cleanup upsampling test * add doc * fix lint * fix lint * fix lint * remove unused import * remove skimage dependency * remove skimage import * remove schedule_upsampling
* add upsampling cpu op * add upsampling gpu schedule * add doc for upsampling op add more doc * cleanup upsampling test * add doc * fix lint * fix lint * fix lint * remove unused import * remove skimage dependency * remove skimage import * remove schedule_upsampling
Only nearest neighbor upsampling is supported.
I think bilinear upsampling can be done by conv2d_transpose with certain weight.