Classification of Trash and Valuables with Machine Vision in Shared Cars
Round 1
Reviewer 1 Report
This paper considers classifying leftover items in shared vehicles into three categories: empty, trash, valuable. The dataset (roughly 2000 images) is collected by the authors using a camera installed in a car. The prediction model is VGG16. Comparisons on prediction accuracies and processing time are made.
The main differences with existing studies are: (a) images used here vary in lighting and shadows, and contain handpicked items; (b) the purpose here is to classify the images into three types (empty, trash, valuable), rather than to recognize specific leftover items.
The following issues need to be dealt with:
(1) equations, say (1) and (2), should be in math mode
(2) line 81, "study[10]" should be saparated by a space
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
- Why the VGG-16 model was adopted instead of some other more advanced deep learning model?
- Due to the small number of samples in the dataset, it is uncertain that the advantages of deep learning can be exerted.
- According to the results in Table 3, the method [2] performs worst while the method [10] performs best. The reasons should be explained. Details of the execution of all these comparative experiments should be given.
- Figure 7 is confusing. The validation set loss oscillates heavily while the validation set accuracy remains the same.
- The training details of the model should be reported, including hyperparameters such as learning rate.
- As we all know, one of the most important things of the successful applications of deep learning models is to improve the generalization ability. Therefore, more related work on regularization and generalization of deep learning models should be discussed or analyzed, including “A full stage data augmentation method in deep convolutional neural network for natural image classification,” Discrete Dynamics in Nature and Society, pp. 1-11, 2020. “Deep facial diagnosis: deep transfer learning from face recognition to facial diagnosis.” IEEE Access 8 (2020): 123649-123661. The successful generalization of deep learning models has brought key help to the task of feature learning and modeling. We believe that the discussion of related work on enhancing the generalization and applicability of deep learning can help to improve the depth of this paper.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
Manuscript ID: applsci-1709865
Type of manuscript: Article
Title: Classification of Trash and Valuables with Machine Vision in Shared Cars
Authors: Nilusha Jayawickrama *, Risto Ojala, Jesse Pirhonen, Klaus Kivekäs, Kari Tammi
Interesting and thorough study but in my opinion with limited impact since a large portion of leftovers and garbage in cars remain in parts that are not visible like compartments, below seats, etc. Also, the authors confused garbage with the cleanness of a car cabin which are different concepts. Also, valuables can be in a huge number of shapes. Just consider different shapes of toys so this study seems rather limited. It is good to point out that the authors made available datasets for further comparison and analysis. Manuscript is clearly written and can be accepted.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 4 Report
The topic is interesting and the manuscript is well composed generally.
1. When setting the three possible output classes, should all bottles be considered as trash? What about the reusable personal bottles?
2. Labels of sub-figures can be used when discussing Figures 4, 6, 9, 10 and 11.
3. There is a format issue in Line 399.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
All the comments have been well revised.
