Unexpected performance comparison between vgg16 and vgg19 for segmentation

Please provide the following information when requesting support.

• Hardware : not relevant
• Network Type: Segmentation
• TLT Version : docker_tag: v3.0-py3
• Training spec file : not relevant
• How to reproduce the issue ?
run ngc registry model list nvidia/tlt_semantic_segmentation:*

The result :

Question : is it normal that the VGG16 has better accuracy and is larger in size/memory foot print than VGG19, or is it just a typo (maybe inverted rows/model names) ?

Thanks, I will check internally.