[Completed][PyTorch] Add Anchor Boxes #1312
Conversation
|
Job d2l-en/PR-1312/7 is complete. |
|
Job d2l-en/PR-1312/17 is complete. |
|
Job d2l-en/PR-1312/35 is complete. |
|
Job d2l-en/PR-1312/36 is complete. |
|
Hi @AnirudhDagar Can you please review the changes made as suggested by you. Thanks |
There was a problem hiding this comment.
Thanks for the PR @suvaansh! Overall LGTM!
Made a few comments PTAL and update the PR.
When you finish a section, just save the notebook.
then run
d2lbook activate all chapter_computer-vision/anchor.mdd2lbook build lib
This will activate all the tabs and then copy the saved functions to the respective libraries.
You need to commit and push updatedd2l/torch.pyfile along with this markdown file.
| def bbox_to_rect(bbox, color): | ||
| """Convert bounding box to matplotlib format.""" | ||
| return d2l.plt.Rectangle(xy=(bbox[0], bbox[1]), width=bbox[2]-bbox[0], height=bbox[3]-bbox[1], fill=False, edgecolor=color, linewidth=2) |
There was a problem hiding this comment.
This method is defined in bounding-box.md
We save it there and use it here.
Similarly for other methods.
| ```{.python .input} | ||
| #@tab pytorch | ||
| #@save | ||
| def multibox_prior(feature_map_sizes, sizes, aspect_ratios): |
There was a problem hiding this comment.
Here it is right that you define this function because torch/torchvision don't offer a built-in method for multibox_prior like mxnet.multibox prior
| #@save | ||
| PredBoundingBox = namedtuple("PredBoundingBox", ["probability", "class_id", "classname", "bounding_box"]) | ||
|
|
||
| def non_max_suppression(bounding_boxes: list, iou_threshold: float = 0.1) -> list: |
There was a problem hiding this comment.
nms not possible through torchvision.ops?
|
This PR has become a mess. So, I am creating a new PR for the same and closing this. It contains all the changes suggested above. Thanks |
Description of changes:
By submitting this pull request, I confirm that you can use, modify,
copy, and redistribute this contribution, under the terms of your
choice.