Tlt 3.0 supports training on multiple resolutions

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

I know that, The tlt-train tool does not support training on images of multiple resolutions, I would like to know whether it supports in tlt 3.0. Or do we need to resize before feeding to the model.

Also can I use the augmentation configuration in spec file to resize my dataset on the run time.

2

In TLT 3.0-dp-py3 version, only detectnet_v2 ,faster_rcnn, Unet need resizing images offline. Other networks do not need to resize images/labels.
In latest TLT 3.0-py3 version , only detectnet_v2 needs resizing images offline. Other networks do not need to resize images/labels.

1 Like
3

Thank you @Morganh

4

Update:
In latest TLT 3.0-py3 version , detectnet_v2 does not need to resize images offline. TLT provides below ways for end user.
See https://docs.nvidia.com/tlt/tlt-user-guide/text/object_detection/detectnet_v2.html#input-requirement

The train tool does not support training on images of multiple resolutions. However, the dataloader does support resizing images to the input resolution defined in the specification file. This can be enabled by setting the enable_auto_resize parameter to true in the augmentation_config module of the spec file.