Add examples of Multi-weight support + model usage #6013
Conversation
datumbox
changed the title
| from torchvision.io.video import read_video | ||
| from torchvision.models.video import r3d_18, R3D_18_Weights | ||
| vid, _, _ = read_video("test/assets/videos/v_SoccerJuggling_g23_c01.avi") |
There was a problem hiding this comment.
@bjuncek Apparently this emits a vague deprecation warning (without a fixed removal date). Does it make sense to remove the warning given that we are still uncertain on what we will do with the specific API? Or perhaps I should modify my example to avoid the warnings?
There was a problem hiding this comment.
Yeah, feel free to remove it.
I'm in the process of re-writing the video-module to have a more sensical stucture
There was a problem hiding this comment.
Thanks @datumbox I made some minor suggestions, LGTM regardless.
I think the snippets describing the different usages are useful, but I wonder if it would make more sense to move them into a gallery example (maybe exampleS)?
Definitely fine as-is though, we can improve on that later (potentially after the release) if we have more bandwidth.
|
|
||
| | | ||
| Here is an example of how to use the pre-trained quantized image classification models: |
There was a problem hiding this comment.
I wonder if we should just refer to the snippet above instead, and mention to pass quantize=True? This seems to be the only difference, apart from the imports.
There was a problem hiding this comment.
I'm OK giving a second example given we have a different namespace. Hopefully on the future this will be deprecated and move all quantization of models on main model builders (with FX quant)
| colors="red", | ||
| width=4, font_size=30) | ||
| im = to_pil_image(box.detach()) | ||
| im.show() |
There was a problem hiding this comment.
I'm getting
RuntimeError: "qnms_kernel" not implemented for 'Float'
But I haven't compiled or updated the nightlies recently, maybe that's why
There was a problem hiding this comment.
That's strange. It works for me. Can you put the recent in case this is a bug?
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>
Summary: * Adding code examples for image classification + quant * Adding code example detection * Adding code example segmentation * Adding code example for video classification * Adding information on how to use the new API. * Putting back the comma. * Apply suggestions from code review * Remove output to avoid staleness from flakiness. * Minor fixes. Reviewed By: datumbox Differential Revision: D36413367 fbshipit-source-id: 661fe7364bc55f7fb1ed1d15c9c72618f03b9177 Co-authored-by: Nicolas Hug <contact@nicolas-hug.com> Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>
4 participants
Fixes partially #6014
Demo link: https://output.circle-artifacts.com/output/job/eb77e259-c40e-488d-b5f5-b3adc3121885/artifacts/0/docs/models_new.html
Multi-weight API example:
Model usage example: