Nikita Ermolenko
Nikita Ermolenko
Test Multik ndarrays as a representation of tensors for Android preprocessing
Add NNAPI execution provider for Android and test it using some compatible model on the device.
Perhaps, in the future we should create a separate module with GPU tests and examples, so there would not be any need to uncomment anything to run them. _Originally posted...
Interestingly, calling `initInputOutputInfo` twice would not change the shapes, because there are checks to initialize them twice. And we need the check for the input shape because for some models...
make drawDetectedObjects method and other visualization methods support BufferedImages as input or [0-255] valued arrays
Currently, if an onnx model requires a 3-channel image but the input provided has only one channel internal onnx library exception will be thrown. ai.onnxruntime.OrtException: Shape [1, 512, 512, 3],...
I think it would be a good idea to unify this (predictSoftly) method with `predictRaw` (as a separate PR). _Originally posted by @juliabeliaeva in https://github.com/JetBrains/KotlinDL/pull/356#discussion_r852313090_
I have used this inspection profile ``` ``` Many issues were found, but not all of them looked useful, so I fixed only part of the warnings. Fixes mostly fall...
We need to set up an infrastructure to run android specific tests and cover existing android specific functionality with tests.
Research options to add efficient conversions from YUV_422_888 format to ARGB_8888 bitmaps