Disable/customise augmentation for the object_detector task #5367
Labels
platform:python
MediaPipe Python issues
stat:awaiting googler
Waiting for Google Engineer's Response
task:object detection
Issues related to Object detection: Track and label objects in images and video.
type:feature
Enhancement in the New Functionality or Request for a New Solution
type:modelmaker
Issues related to creation of custom on-device ML solutions
MediaPipe Solution (you are using)
Object Detection (Model Maker)
Programming language
Python
Are you willing to contribute it
No
Describe the feature and the current behaviour/state
Currently, while the Image Classification task has "do_data_augmentation", the Object Detection Task does not. The object detection task should be provided this option.
It would also be useful if both could allow for customisation. TF-Vision exposes aug_scale_min, aug_scale_max, aug_rand_hflip, aug_rand_vflip, and aug_type; where aug_type can take a tfm.vision.configs.common.Augmentation (i.e. AutoAugment or RandAugment)
Will this change the current API? How?
No response
Who will benefit with this feature?
No response
Please specify the use cases for this feature
My dataset includes "right arrow" and "left arrow". Random Flipping is enforced and cannot be disabled, without subclassing the object detection class. This means the classes cannot be trained.
Any Other info
No response
The text was updated successfully, but these errors were encountered: