Koan-Sin Tan

Results 32 issues of Koan-Sin Tan

some ideas we can try to improve the apple backend - [ ] in the WWDC 2023 "Use Core ML Tools for machine learning model compression", https://developer.apple.com/wwdc23/10047, Apple folks claimed...

priority:low

- Open source third party libraries - TensorFlow: we use TFLite (the default fallback runtime) and some TensorFlow components (for preprocessing and evaluation). Is TensorFlow Apache 2.0 enough? Seems not...

Let's check what need to be done before submitting to App Store - [ ] screenshots - [ ] requirement 2 And then submit it.

1. host machines and all the supported devices... 2. air-conditioned? 3. manually intervention??

Currently, we have to manually enable the "Manage All Files Access Permission". That's not intuitive and could not be submit to Google Play Store. https://github.com/mlcommons/mobile_app_open/pull/621

@anhappdev please gather related issues there

- [x] EULA, license stuff - #873, @RSMNYS do we have the consolidated licenses file? - [x] update the "About" page, it doesn't show latest SoCs. #892 - [x] Any...

let's check all other possible options https://github.com/mlcommons/mobile_app_open/issues/840. For 4.0 to the Play Store, we can just remove the file permission and use `/sdcard/Android/.....`

Currently, image preprocessing in `tensorflow/lite/tools/evaluation` only supports `jpeg` and `raw` images. This patch add `png` support. Why we need `png`: 1. raw rgb files are larger 2. the jpeg used...

awaiting review
comp:lite
size:M

the tflite implementation https://github.com/mlcommons/mobile_app_open/blob/master/mobile_back_tflite/cpp/backend_tflite/stable_diffusion_pipeline.h#L40 supposed, both - the seed, and - the number of steps (default to 20) should be configurable from the app or from the pbtxt (task description,...