Hello, Thank you for your impressive work and for releasing the code.
I have a question regarding Table 5. I am currently trying to reproduce the reported inference time following the same experimental setting. However, in my experiment using UniDepthV2-old at the same input resolution (228×304), a single test-time optimization iteration takes approximately 0.595 seconds. As a result, performing 50 iterations, as described in the paper, requires more than 30 seconds in total.
I would like to ask whether there might be any specific implementation detail or experimental setting that I may have missed (e.g., feature reuse, reduced optimization steps, or different timing protocol) when measuring the inference speed reported in Table 5.
Thank you very much for your time and clarification.
Best regards,
[I ran both the official code provided by the authors, and UniDepthV2-old official code]
Hello, Thank you for your impressive work and for releasing the code.
I have a question regarding Table 5. I am currently trying to reproduce the reported inference time following the same experimental setting. However, in my experiment using UniDepthV2-old at the same input resolution (228×304), a single test-time optimization iteration takes approximately 0.595 seconds. As a result, performing 50 iterations, as described in the paper, requires more than 30 seconds in total.
I would like to ask whether there might be any specific implementation detail or experimental setting that I may have missed (e.g., feature reuse, reduced optimization steps, or different timing protocol) when measuring the inference speed reported in Table 5.
Thank you very much for your time and clarification.
Best regards,
[I ran both the official code provided by the authors, and UniDepthV2-old official code]