Use Pothole Dataset in Tutorial for AutoMM Detection #2468
Conversation
FANGAreNotGnu
changed the title
|
Job PR-2468-8750045 is done. |
docs/tutorials/multimodal/object_detection/data_preparation/prepare_pothole.md
Show resolved
Hide resolved
|
Job PR-2468-c376d20 is done. |
|
Job PR-2468-75b17dc is done. |
|
I'm previewing the content via http://autogluon-staging.s3-website-us-west-2.amazonaws.com/PR-2468/75b17dc/tutorials/multimodal/object_detection/finetune/detection_fast_finetune_coco.html and I feel that you can put the label also in the starter image. |
|
Also, for the figure in http://autogluon-staging.s3-website-us-west-2.amazonaws.com/PR-2468/75b17dc/tutorials/multimodal/object_detection/finetune/detection_high_performance_finetune_coco.html#sec-automm-detection-high-ft-coco , you may like to also attach the ground-truth of the test-image. |
|
The tutorial looks much better now! |
|
Job PR-2468-8eafacc is done. |
|
Job PR-2468-b065f41 is done. |
There was a problem hiding this comment.
The tutorial looks great. I think the comment from @sxjscience is to add GT bboxes on the first image as well. In that case, (1) people will immediately know the task is to detect these holes (2) they can compare to our predictions in the end to see how the detector works.
|
The GT bboxes are added to the sample image. An interesting thing is that vfnet-r50 detects more (reasonable) potholes than in groundtruth. |
|
Job PR-2468-6906000 is done. |
By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.