Research on Computer Vision models

[Lukasnol]

Detailed Description

What can previous models recognise and how well?
What other classifications are there in the models?
Are there newer versions of the models in pytorch / ultralytic?
Should I use several models at the same time?

Definition of Done

• see description above

Effort Estimate

No response

Testability

No response

Dependencies

No response

[ll7]

I just realized that @MaxJa4 asked himself a similar question last semester. There should still be an evaluation script for it somewhere in the repo, if I'm not mistaken. He also published the results on the project's Google Drive folder. Do you know how to get access to the Google Drive folder?

[Toni2go]

I don't know how to access the Google Drive folder to be honest.

[ll7]

There is a pinned message on our discord server. You are welcome to contact me if you experience any issues.

[SirMDA]

@Toni2go With the current research about the intermediate-layer it would be nice to know if there is a possibility to get a model which makes object detection on the lidar data. If this issue is addressed research about this idea would be appreciated. This would have the advantage of detecting objects not only in front of the car but also around it without adding another sensor.

[Toni2go]

@SirMDA I did something like that in my Bachelor thesis actually. But it was with a 2D Lidar sensor and I programmed the algorithm in C, so with some adjustments I think I know how to solve this (with a 3D Lidar in Python). At least if it's only interesting to know if there is something in the way. My idea would be to represent the surrounding obstacles with a line. Do you think that would be applicable?

[ll7]

@Toni2go with your experience in point cloud object detection, could you decide whether this project https://github.com/maudzung/SFA3D is worth investing any time? Maybe we should keep @asamluka in the loop.