HDR-Transformer

HDR-Transformer copied to clipboard

Thank you for your great work.

In your paper, in Table 2 you report an inference time for your proposed network of 0.15s, which is twice as fast as previous state of the art model HDR-GAN.

Using a single Quadro P6000 I obtain an inference time of 6 seconds for HDR-Transformer using your MegEngine repository and 7 seconds using your Pytorch version while I'm able to reproduce the inference time for other networks.

Can you give more information on how to achieve this inference time? Which hardware is used? Do you use post-training optimisation?

Thank you