Jianing Zhang

@buntupana i tried what you said by using CameraView to record videos and Mp4Parser to append them together, i still get second video that to be concatenated upside down, i...

Hi @HBiSoft , i just ran into the same issue again. Tested with my own mp4 video and @reijin90 video, both of them throw me IndexOutOfBoundException ~~~~java java.lang.IndexOutOfBoundsException: index=0 out...

After some researching, i found a work around. If you have a mp4 video, try install ffmpeg CLI tools, for mac users, `brew install ffmpeg`. Once it's installed, then run...

Hi @sgowroji, Of course, I will share code changes regarding the holistic graph here, since it's relevant. Noticed that we have our private customised google/mediapipe-based repo, with updating Mediapipe from...

Hi @ivan-grishchenko @sgowroji , do you have time to look at this issue in any possible way? thanks!

Hi @ivan-grishchenko @sgowroji, according to the logs above, it seems that when holistic tracking is activated in 0.8.6 and higher, _ReplaceNodeSubsetsWithDelegateKernels_ in the subgraph.cc was invoked and node replacement sets...

Hi @jiuqiant, appreciated if any tips you could shed some light on this issue

I have tried the latest 0.8.10 build, the problem still persists. @chuoling is there anything you could comment on this issue or give me some guidance or hint? Thanks!

Just an update here, i managed to solve it by using different delegate option. Before, the inference calculator uses the gpu delegate: ``` node { calculator: "InferenceCalculator" input_stream: "TENSORS:input_tensors" output_stream:...

After going through your issue #3545 , think we got different issues. xnnpack is cpu indeed, although gpu delegate supposed to be faster especially on my pixel 5 which supports...