search

dividiti / crowdsource-video-experiments-on-android

Crowdsourcing video experiments (such as collaborative benchmarking and optimization of DNN algorithms) using Collective Knowledge Framework across diverse Android devices provided by volunteers. Results are continuously aggregated in the open repository:
http://cKnowledge.org/repo
Other
28 stars 3 forks source link

strange issue with recognition using AlexNet on ACER B1-780 #65

Open gfursin opened 7 years ago

gfursin commented 7 years ago

I have a strange issue on a tablet ACER B1-780 (Android 6.0, ARM64). Recognition is fine on 3 first models (GoogleNet, SqueezeNet), but fails on CaffeNet and AlexNet. In fact it returns the following error:

bionic_open_tzdata_path: ANDROID_DATA not set! bionic_open_tzdata_path: ANDROID_ROOT not set!

and empty recognition. I don't know if it will be easy to reproduce such issue, because all models work fine on my S7 ...

DVEfremov commented 7 years ago

I've got same warns (not error) on my Galaxy Note 4 but it works fine.

DVEfremov commented 7 years ago

@So after cmd excution on my android I've got stdouts like:

recognitionResult[0] = ""
recognitionResult[1] = "---------- Prediction for /sdcard/openscience/data/4a6e91d75d682cec/demo/mouse1.jpg ----------\n0.9003 - "n03793489 mouse, computer mouse"\n0.0118 - "n03929660 pick, plectrum, plectron"\n0.0081 - "n04317175 stethoscope"\n0.0058 - "n04548280 wall clock"\n0.0058 - "n04286575 spotlight, spot"\n"
recognitionResult[2] = "__bionic_open_tzdata_path: ANDROID_DATA not set!\n__bionic_open_tzdata_path: ANDROID_ROOT not set!\n__bionic_open_tzdata_path: ANDROID_DATA not set!\n__bionic_open_tzdata_path: ANDROID_ROOT not set!\n__bionic_open_tzdata_path: ANDROID_DATA not set!\n__bionic_open_tzdata_path: ANDROID_ROOT not set!\n"

Where output string list:

gfursin commented 7 years ago

Hmm, so it's not problem. But the thing is that I do not get any stdout (it's empty) and no further errors ... Need to check it deeper at some point ...

gfursin commented 7 years ago

I debugged new app on my cheap tablet and all models but AlexNet/CaffeNet work fine, but the latter ones do not work and do not give an error or output, though program clearly do something for about 7 secs... What I think is maybe there is not enough memory (these cheap devices has 1Gb of memory with 4 cores) while error just suppressed somewhere in classification bin ... Strange ... Anyway, we can leave it for now, or we can try to run it on a similar device with 1Gb of memory and later debug classification bin itself ...