Occlusion Avoidance in a Simulated Environment Using Reinforcement Learning
Mir Reza Ghaffari Razin
Round 1
Reviewer 1 Report
0. Do the authors meet with occluded-coco dataset for object detection task https://paperswithcode.com/dataset/occluded-coco ? and work in this field?
1. line 23: in real-world situations. What kind of situations, authors means?
2. line 37: "The focus of our paper is partial occlusion, as it is one of the most significant and common sources of disturbance." - please describe some case study of the cases where this is important
3. line 43: "as that allows us to create high-quality labelled datasets with minimal costs." - what the labelling type assumed here , please also clarify.
4. The introduction contains too small number of references.
5. after line 59 authors need to suggest some examples of obj.det tasks with their description or references where the problem of occlusions is met.
6. contribution in the line 69: i do not see any relation to the object detection task.
7. line 106 what authors means " occlusion labels "
8. liens 98-103 may I recommend authors shows their knowledge with SOTA in the investigated field by citing some surveys or at least https://paperswithcode.com/sota/object-detection-on-coco and https://paperswithcode.com/sota/real-time-object-detection-on-coco pages.
9. "Still, none of these detection methods address the problem of input disturbances 104 (especially partial occlusion) directly. " - The object detection methods
10. line 325 - at least put a reference to U-Net and SR-UNet paper. ALso the architectures, descriptions in the sense of research should be given.
11. line 325 "Both the encoder and decoder subsystems are designed as U-Net-like structures. We designed and trained numerous different variations of these parts" - U-Net is combined approach including both encoder and decoder
12. line 320: please clarify: "is capable of producing the occlusion masks as well as 320 the gradients."
13. please describe the relation of section 5 with the object detection task.
Author Response
Dear Reviewer!
Thank you for your valuable input, please see our responses in the attached pdf.
Best regards,
Mátyás Szántó
corresponding author
Author Response File: 
Author Response.pdf
Reviewer 2 Report
To further improve the current state of the work, a dataset section should be introduced with additional pseudocode on how it was generated. The authors should also highlight the dataset's uniqueness and how it differs from the existing dataset.
Author Response
Dear Reviewer!
Thank you for the valuable points you raised. Please see our responses attached.
Best regards,
Mátyás Szántó
corresponding author
Author Response File: Author Response.pdf
Reviewer 3 Report
if possible author can explain his mathematical model with equations
explain the limitations of the paper in the conclusion section
is this model is only for synthetic environment, how it behaves for complex or real time environment
when you talk about camera parameters, can you make it clear weather it is intrinsic or extrinsic or both
talk about why you used pytorch, what processor you used as sub-paragraph
Author Response
Dear Reviewer!
Thank you for the valuable points you raised. Please see our responses attached.
Best regards,
Mátyás Szántó
corresponding author
Author Response File: Author Response.pdf
Reviewer 4 Report
1-The numerical and statistical results of the article must be mentioned in the abstract.
2- Edit the introduction and refer to newer papers. Also, the results of some new papers should be mentioned.
3- The mathematical theory of the paper should be presented more clearly and in more detail.
4- Figures and numerical results should be presented with clearer interpretation.
Author Response
Dear Reviewer!
Thank you for the valuable points you raised. Please see our responses attached.
Best regards,
Mátyás Szántó
corresponding author
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
To some degree, the authors addressed all my responses.
