experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Update person detection README to reflect new cmsis-nn optimized inference speeds.

Browse Source 
PiperOrigin-RevId: 360460921
Change-Id: I7f030bf2633844dba320609cb5a95ee1fecdf964
This commit is contained in:
Nat Jeffries 2021-03-02 10:31:05 -08:00 committed by TensorFlower Gardener
parent 8156c22727
commit 7e55a20063
1 changed files with 7 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
tensorflow/lite/micro/examples/person_detection/README.md
View File

@ -227,18 +227,17 @@ build and upload the example.
To test the camera, start by pointing the device's camera at something that is
definitely not a person, or just covering it up. The next time the blue LED
flashes, the device will capture a frame from the camera and begin to run
inference. Since the vision model we are using for person detection is
relatively large, it takes a long time to run inference—around 19 seconds at the
time of writing, though it's possible TensorFlow Lite has gotten faster since
then.
inference. The vision model we are using for person detection is relatively
large, but with cmsis-nn optimizations it only takes around 800ms to run the
model.
After 19 seconds or so, the inference result will be translated into another LED
being lit. Since you pointed the camera at something that isn't a person, the
red LED should light up.
After a moment, the inference result will be translated into another LED being
lit. Since you pointed the camera at something that isn't a person, the red LED
should light up.
Now, try pointing the device's camera at yourself! The next time the blue LED
flashes, the device will capture another image and begin to run inference. After
19 seconds, the green LED should light up!
a brief puase, the green LED should light up!
Remember, image data is captured as a snapshot before each inference, whenever
the blue LED flashes. Whatever the camera is pointed at during that moment is