Week Project-Guava Grading

In this project, We'd like to integrate real time object detection and mobile in practice.

Demo

We may assume that:
1. Generally, My grandfather picks immature fruits at 5am-7am. The dataset was collected without direct sunlight, so the performance may get worse as we use the app at other times.
2. The app is ONLY used on the field.

Guava datasets

There are four labels listed below: 1, 2, immature, and background. "1", "2" are the grading of guava (means level one and level two). There are totol 2244 images in the dataset.

MobileNets

We used the mobilenet in keras. The parameters are:
weights : None
include_top : True
input shape : 224 * 224 * 3
classes : 4
batch size : 10
epochs : 20
validation number ratio : 0.33
ModelCheckpoint : save_best_only=True

Figure