How we can add online transform augmentation to LPRnet wirh TLT3.0?

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

• Hardware (T4/V100/Xavier/Nano/etc) : GTX 1080ti
• Network Type (Detectnet_v2/Faster_rcnn/Yolo_v4/LPRnet/Mask_rcnn/Classification/etc): LPRnet
• TLT Version (Please run “tlt info --verbose” and share “docker_tag” here) : 3.0 last version

The TLT3.0 is great tool for fine-tuning models without any programming codes, but it has one big drawback.
As you know the online data augmentation is a excellent technique for learning model to get best accuracy and robustness.
In the TLT3.0, There isn’t some of great data augmentations, like translate left/top/down in the training LPRNet and detection models, or shearing, …
If it’s possible add new data augmentations like translate into TLT3.0 as online augmentation.

2

There are online data augmentation in object detection networks, such as detectnet_v2, faster_rcnn, ssd, dssd, retinanet, yolo_v3 and yolo_v4.

For LPRnet, there is some augmentation mentioned in NVIDIA TAO Documentation

3

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.