The common object detection accuracy is poor when deployed a default tflite in mobile #20
Description
Installed the default tflite apk file in the mobile device using the android studio environment. By default, the installed app has to detect the 90 common objects. The detecting accuracy is very poor.
System information
I didn't make any changes to the TFLIte framework. I cloned the GitHub repo (https://github.com/tensorflow/examples.git) and opened the object detection folder in the android studio. I generated a build and installed that generated Apk file.
Used windows 10 with android studio
Mobile device (Samsung A30)
TensorFlow installed from (https://github.com/tensorflow/examples.git
TensorFlow version 1.15
Describe the current behavior
After installing the tflite default(90 classes) apk file into the mobile. Checked with various object what mentioned in the labelmap file. But the prediction outcome is not correct most of the time.
Why this poor accuracy with the default TFLite Apk file.