Robotic Inference

Include in your project submission:

• your write up (PDF format)
• any supporting images (including sample images of the items you are classifying, charts, graphs, etc.)
• your trained model for the supplied data set.

For the submission of your trained models include the following files in separate folders:

• deploy.prototxt
• labels.txt
• mean.binaryproto
• your_model.caffemodel
• solver.prototxt
• train_val.prototxt.

Also included, a photo of the evaluate command output as a screenshot with your name annotated on it.

Annotation can be in a variety of ways, but using a watermarking tool such as https://www.watermarquee.com is preferred.

Student submitted all required files specified in criteria.

• PDF format write-up
• Trained model for the supplied data set
• Supporting images (optional)

A network of your choice must be trained on the supplied data set and must fall below the required inference time of 10 ms on the supplied workspace.

Student includes a watermarked screen shot that passes the minimum required inference time (10ms) and has correctly added his/her name. The name should be a watermark that covers the image, but does not make items illegible.

A network of your choice must be trained on the supplied data set and must surpass the required accuracy of 75%.

Student includes a watermarked screen shot that passes the required accuracy of 75% and has correctly added his/her name. The name should be a watermark that covers the image, but does not make items illegible.

Include a full write up with the following sections:

• Abstract
• Introduction
• Background / Formulation
• Data Acquisition
• Results
• Discussion
• Future Work.

Include supporting images where appropriate. All images, charts, tables, etc. must be watermarked in the report.

Student includes a full write up covering the required sections. The write up is of proper and professional formatting with support images where appropriate. The write-up must be submitted in PDF format.

Abstract - Give a high level overview of your work

In the abstract, student gives a high level overview of what is being attempted in the report. Abstracts are typically 5-10 sentences that provide just enough context to understand the gist of the report.

Introduction - Introduce your robotic inference idea

The introduction should provide some material regarding the history of the problem, why it is important and what is intended to be achieved. If there exists any previous attempts to solve this problem, this is a great place to note these while conveying the differences in your approach (if any). The intent is to provide enough information for the reader to understand why this problem is interesting and set up the conversation for the solution you have provided. Use this space to introduce your robotic inference idea and how you wish to apply it. If you have any papers sites you have referenced for your idea, please make sure to cite them.

Background / Formulation - Explain why you chose the network you did for the supplied data set and then why you chose the network used for your robotic inference project.

At the background / formulation stage, you should begin diving into the technical details of your approach by explaining to the reader how hyperparameters for training the network were defined, what type of network was chosen, and the reasons these items were performed. This should be factual and authoritative, meaning you should not use language such as I think this will work or Maybe a network with this architecture is better… Instead, focus on items similar to, A 3-layer network architecture was chosen with X, Y, and Z parameters Explain why you chose the network you did for the supplied data set and then why you chose the network used for your robotic inference project.

Data Acquisition - Why did you choose the data you did for your robotic inference project? How did you collect the data? What are important characteristics to capture in the data?

The Data Acquisition section should discuss the data set. Items to include are the number of images, size of the images, the types of images (RGB, Grayscale, etc.), how these images were collected (including the method). Providing this information is critical if anyone would like to replicate your results. After all, the intent of reports such as these are to convey information and build upon ideas so you want to ensure others can validate your process. Justifying why you gathered data in this way is a helpful point, but sometimes this may be omitted here if the problem has been stated clearly in the introduction. It is a great idea here to have at least one or two images showing what your data looks like for the reader to visualize.

Results - Discuss the results of your robotics project model and the model you used for the supplied data with the appropriate accuracy and inference time.

Results part is typically the hardest part of the report for many. You want to convey your results in an unbiased fashion. If you results are good, you can objectively note this. Similarly, you may do this if they are bad as well. You do not want to justify your results here with discussion; this is a topic for the next session. Present the results of your robotics project model and the model you used for the supplied data with the appropriate accuracy and inference time For demonstrating your results, it is incredibly useful to have some charts, tables, and/or graphs for the reader to review. This makes ingesting the information quicker and easier.

Discussion - Reflect on which is more important, inference time or accuracy, in regards to your robotic inference project.

Discussion is the only section of the report where you may include your opinion. Make sure your opinion is based on facts, in this case, your results. If your results are poor, make mention of what may be the underlying issues. If the results are good, why do you think this is the case? Also, reflect on which is more important, inference time or accuracy, in regards to your robotic inference project.
Again, avoid writing in the first person (i.e. Do not use words like I or me). If you really find yourself struggling to avoid the word I or me; sometimes, this can be avoid with the use of the word one. As an example: instead of, "I think the accuracy on my dataset is low because the images are too small to show the necessary detail" try, "one may believe the accuracy on the dataset is low because the images are too small to show the necessary detail". They say the same thing, but the second avoids the first person.

Conclusion / Future Work - How will you or did you leverage embedded workflow for personal growth in robotics? Would this be a commercially feasible product? Why or why not?

The Conclusion / Future Work section is intended to summarize your report. Your summary should include a recap of the results, did this project achieve what you attempted, and is this a commercially viable product? For Future work, address areas of work that you may not have addressed in your report as possible next steps. For future work, this could be due to time constraints, lack of currently developed methods / technology, and areas of application outside of your current implementation. Again, avoid the use of the first-person.