search

ilastik / lazyflow

lazy parallel ondemand zero copy numpy array data flows with caching and dirty propagation
Other
77 stars 59 forks source link

Enable constant BlockShape requests for NN classification #284

Closed FynnBe closed 5 years ago

FynnBe commented 5 years ago

Tiktorch networks expect valid shapes. We cannot meaningfully extend unsuitable requests on the Tiktorch side, as the raw data resides elsewhere (atm with ilastik).

In the current version of the nn workflow these ops are responsable for relevant block shapes:

This should be extended to a list of valid block shapes for better performance, but these are only the edge cases.

FynnBe commented 5 years ago

The caches/BigRequestStreamer should operate with the max block shape. Smaller prediction requests are handled by the classifier: ilastik/lazyflow@31c7a905fd692b36084c19c314932d170de4fe89