-
Notifications
You must be signed in to change notification settings - Fork 862
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
[multimodal]Upgrade object detection backend to mmdet 3.0 #3188
[multimodal]Upgrade object detection backend to mmdet 3.0 #3188
Conversation
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
Thanks for updating the version. Can you provide some benchmark results and attach them in the description? Thanks |
Job PR-3188-7cd1dc2 is done. |
Just completed the basic changes. I'll run a small benchmark after working on a PR in autogluon-bench. And will do a complete benchmarking after upgrading presets to more recent models (RTMDet/DINO). |
Job PR-3188-50bedc8 is done. |
Job PR-3188-70feb9d is done. |
282961f
to
d7e99b5
Compare
Job PR-3188-78157d1 is done. |
Job PR-3188-0635abd is done. |
...utorials/multimodal/object_detection/finetune/detection_high_performance_finetune_coco.ipynb
Show resolved
Hide resolved
multimodal/src/autogluon/multimodal/configs/pretrain/detection/yolox/yolox_l_8xb8-300e_coco.py
Show resolved
Hide resolved
multimodal/src/autogluon/multimodal/data/process_mmlab/process_mmdet.py
Outdated
Show resolved
Hide resolved
return rets | ||
elif mode == "predict": | ||
# for detailed data structure, see https://github.com/open-mmlab/mmdetection/blob/main/mmdet/structures/det_data_sample.py | ||
return [{BBOX: ret.pred_instances, LABEL: ret.gt_instances} for ret in rets] |
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.
Why does predict
mode require groundtruth labels?
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.
model(mode="predict")
is used for both predicting and evaluation. It's easier to include the ground truth here than merging ground truth with result in lit module / predictor. Here gt_instances
will be None
if there is no label available.
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.
The evaluation is done after prediction. Why do we need labels in the prediction?
multimodal/src/autogluon/multimodal/configs/model/fusion_mlp_image_text_tabular.yaml
Show resolved
Hide resolved
Job PR-3188-0c266ee is done. |
Job PR-3188-c00a1e1 is done. |
Job PR-3188-f54327f is done. |
multimodal/src/autogluon/multimodal/data/process_mmlab/process_mmlab_base.py
Outdated
Show resolved
Hide resolved
multimodal/src/autogluon/multimodal/data/process_mmlab/process_mmlab_base.py
Outdated
Show resolved
Hide resolved
Job PR-3188-5990923 is done. |
Job PR-3188-e2ee022 is done. |
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.
LGTM. Thanks for the upgrade.
Change MMDetection backend to 3.0.
Changes:
Benefits:
Future works:
By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.