The code that you submitted to the Fishyscapes submission?

[pswena]

I saw the issue 13, but could you please upload your code which you submitted to the Fishyscapes submission to get the result of table 1? Please.

[hermannsblum]

Sure, you can find that code here: hermannsblum/fishyscapes#4

[pswena]

Sure, you can find that code here: hermannsblum/fishyscapes#4

Thank you very much! Thank you very much! Cause I am a freshman in deep-learning, so I asked lots of questions. Thank you for you reply.

[pswena]

Firstly, I can't find test_fishy_torch in driving_uncertainty. Is that you missing this file?
Secondly, I want to replicate the result of table 1, should I submit the code you mentioned before to the Fishyscapes submission?

[hermannsblum]

I may miss the point, but you cannot replicate the result of table 1 yourself. The results for Synboost are available at https://fishyscapes.com/results . Table 1 is literally copy-pasted from there.

[pswena]

I just want to replicate your results of table 1 in FS Lost & Found 、FS Static and FS Web datasets. So what should I do? Which code should I submit to the Fishyscapes submission? Or, I cannot replicate by myself.

[hermannsblum]

No, you cannot replicate this by yourself. The idea of public benchmarks and challenges is that their test-sets are hidden and only they can measure/evaluate your models, such that you cannot overfit on the test sets.

[pswena]

No, you cannot replicate this by yourself. The idea of public benchmarks and challenges is that their test-sets are hidden and only they can measure/evaluate your models, such that you cannot overfit on the test sets.

So, you mean I have no idea to replicate the result that you mentioned in your paper, right?

[CesarCadena]

@pswena you always have the validation sets available for you to replicate results (for example table 2 in the paper).

[pswena]

Yeah, I have done it. Thanks for your replay. But I have another question. Can I run your code on FS Lost & Found 、FS Static and FS Web datasets? Not to replicate the result, just curious about how to run the code on Fishyscapes benchmark.

[hermannsblum]

If you want to try it out on some additional images, I would recommend these: https://segmentmeifyoucan.com/ The test-images in that benchmark are available for download, so you can run Synboost over them.

[pswena]

If you want to try it out on some additional images, I would recommend these: https://segmentmeifyoucan.com/ The test-images in that benchmark are available for download, so you can run Synboost over them.

Thanks for your quick responses and congratulations for the paper! Thank you very much.

[pswena]

I may miss the point, but you cannot replicate the result of table 1 yourself. The results for Synboost are available at https://fishyscapes.com/results . Table 1 is literally copy-pasted from there.

May I ask you for the test_fishy_torch in driving_uncertainty?