Hi,

I haven't noticed that before merging, but the way this DelayedSamplesCall handles data is not super convenient in terms of memory usage, don't you think? https://gitlab.idiap.ch/bob/bob.pipelines/-/blob/master/bob/pipelines/wrappers.py#L57 Data are stacked at the very beginning of the pipeline, which can be huge in size (e.g raw images/video).

The way we had it before !26 (merged), was more convenient. This stacking was done only when necessary (once data is "more preprocessed").

Is there any reason to be like that?

Thanks