WASTE SORTING /DESIGN + RESEARCH

Reflection for part 1
1.After training, model can be used for waste sorting. It can detect the random images
2.The accuracy of detecting result still needs to be improved. It cannot give a accurate detections for those items which have confusing appearance
3.100 images for each object may be not enough to give accurate detections. More images with different perspective and background are needed for a better accuracy
4.Training time is crucial dilemma for model training. Much more time will be needed for more images and more categories.

Application demonstration on AVD

Reflection for part 2
1.For an application of waste sorting, 50 objects are not enough. More objects are needed to be trained
2.To improve the accuracy of detection result, more training images are needed, which means much more time will be needed for training.
3.Detection result may be more confusing when more objects are added for detection.
4.The running speed on android virtual device is quite slow and there are some delays. Problem may occur by overcomplicated codes or the large trained data.