hows the inference speed compares to monodepth2 in terms of same accuracy?

[lucasjinreal]

hows the inference speed compares to monodepth2 in terms of same accuracy?

[ranftlr]

Not sure how to answer this question as there is no "same accuracy". These are different models that have been trained in a different way and as such will always have different accuracy on any given dataset.

As both are freely available, I suggest to just run both yourself and see what works better for your use case.

[lucasjinreal]

@ranftlr I mean closely performance, say mAP both 35-36, and compare the speed. I am not sure if DPT can run efficient. If it slower and not more accurate than monodepth2, then I should not try DPT. (not disavowal DPT, just prefer more fast and accurate model)

[ranftlr]

Ok, feel free to (or also not to) try and judge yourself. Setup of DPT takes about a minute.

[lucasjinreal]

@ranftlr Do u think DPT is faster and accurater?

[phamdat09]

@jinfagang As I tested, DPT is much accurate than monodepth2. But the speed is slower.
Just see the size of 2 models, I can guess the speed of both

[pengzhi1998]

@phamdat09 Sorry to bother you. But how long will it take for DPT-large to estimate an image's depth value?

[phamdat09]

@phamdat09 Sorry to bother you. But how long will it take for DPT-large to estimate an image's depth value?

as I tested with Tensorrt avec Rtx2080. FPS is around 20 including pre-processing and post-processing

[pengzhi1998]

@phamdat09 Thank you!