Hello,
I have a question regarding the FPS metrics reported in the your paper.
Upon reviewing the CustomSemSegTester class, I noticed that the inference speed is calculated differently in two functions:
test(): Measures time taken per entire frame.
test_one_frame(): Measures time taken per fragment.
Could you please clarify if the FPS values for PTv3, PTv3-distilled, and all other baseline models in the paper were measured based on the test_one_frame (per-fragment) logic?
Thank you for your time and for this great work!
Hello,
I have a question regarding the FPS metrics reported in the your paper.
Upon reviewing the CustomSemSegTester class, I noticed that the inference speed is calculated differently in two functions:
test(): Measures time taken per entire frame.
test_one_frame(): Measures time taken per fragment.
Could you please clarify if the FPS values for PTv3, PTv3-distilled, and all other baseline models in the paper were measured based on the test_one_frame (per-fragment) logic?
Thank you for your time and for this great work!