Sacha

Introduce TARGET fallbacks in CMake

12

X86(-64) targets for cross-compilation are likely to be for a simulator/testing and not a production environment. Regardless, we do not know information about the target yet, so choose a default...

https://github.com/cavalleria/cavaface.pytorch/blob/e9b9bd8ee06de51649ee202712d673f8e64415e9/backbone/resattnet.py#L79 Is it intentional that the stage1 attention block uses addition for the out_trunk whereas the rest use multiplication? Other repositories that implement this method appear to use multiply here,...

Suggestion: Use Nvidia DALI as the dataloader

4

Would improve the input speed.

Requires width factor of 1.5 to be correct size for input

Model for LFW

1

Which model did you use to achieve 99.85% on LFW? Is it LightCNN?

Allow user to switch between yuv420 and nv12

1

It was hardcoded to use yuv420. User setting to anything else would break it.

Allow NV12 choice

2

`+ nvmpi_context->ctx=nvmpi_create_decoder(codectype,NV_PIX_YUV420);` In the patch, it is hardcoded to use YUV420. Is there a reason why this is preferred over NV12? Edit: I tried replacing YUV420 with NV12 and received...