Questions about evaluation metrics and experimental results

[yoookoo]

• Latency is relevant to the evaluation equipment, so why do not you report Parameters and FLOPs? And the code is not available.

• In your demo [https://hubert0527.github.io/InstaNAS/], networks are wired in the same architecture when facing different input samples. I do not see the input-dependent form. It seems your method learned a fixed architecture in your distribution, which is a conflict with your motivation.

• So I doubt the fairness of the experimental results.