Jian Shi

Results 72 comments of Jian Shi

The ```preprocessing``` and ```postprocessing``` should happen in containers like an ```ImageSequential```. For the ```Algorithm``` part, I would like to see it happen. With a clean usage like: ```python method =...

> Is there any minimal test that we can add here ? Probably. I will investigate what are worth testing.

Can you elaborate what are the missing pieces in your experiment? Also, we are working on supporting the detection now. https://github.com/kornia/kornia/pull/1497 /cc @twsl

@twsl Great. If possible, may you share the demo after #1497 get merged?

@tileb1 Thanks for reporting. I will fix it this weekend. The documented methods shall work with earlier version like 0.6.0, before the refactor of the new random generator.

@tileb1 Can you play with #1675 to see if it works to your case? Please consider to contribute with several testing cases.

@tileb1 Can you try using `cropping_mode=='resample'?`

> One should never use other interpolation methods other than `nearest` for mask, or you'll end up with non-binary values. I am not against making different default flags for different...

I am thinking to introduce enums like MASK.INTEGER and MASK.FLOAT for this. So that the interpolation method will be set to nearest when the input is MASK.INTEGER. MASK.FLOAT will maintain...

> Disregarding for now the fact that masks and inputs should not be treated the same, how could we allow augmentations without DataKey.Input? It seems the only reason to force...