Tune SCARF parameters before trying to fine-tune the network as it's easier and less computation costly
Use RBG images from which you can extract edges to tune the model as they are very similar to SCARF representation
Modify the introduction adding drawbacks from the cited previous methods, as to highlight the motivation for why this project is interesting
Add the conclusion also adding some use-case scenario. I.E.: introduce EDOPT and the initialization problem and highilight that with my work an improved pipeline can be created allowing to solve the initialization problem in which if the object isn't in the correct position it can be detected thanks to this network
Add a slide in which to explain the main motivations of this research
Show results (when I will have some)
Highilight that it's not really a copy of OnePose with some improvement, as the networks will be very different as they have a different setting
Link to 1st presentation
Feedbacks: