Create a script to produce results

[DanielCoelho112]

Right now we have a script to collect the data, another to train the model, and now we need one to produce results.

This script must receive as input one dataset and one model. The output should be a file/folder that contains both the predicted poses and the real poses. I don't know if the output should be a file, for example a .csv, or a folder with the poses as .txt files as the ones in the dataset.

This file/folder will then be the input of #15.

We should discuss this tomorrow @miguelriemoliveira @pmdjdias

[DanielCoelho112]

Hi @miguelriemoliveira and @pmdjdias,

Check the results folder:

The folder contains the following information:

• Predicted and real pose for each frame
• csv file with the position and rotation error for each frame, and the average values
• config file with information regarding the dataset and model used.

[miguelriemoliveira]

Hi @Daniel Coelho @.***>

Nice work. The csv file should have the units.

Is it meters? If so that's a lot right?

On Sat, Mar 26, 2022, 16:24 Daniel Coelho @.***> wrote:

Hi @miguelriemoliveira https://github.com/miguelriemoliveira and @pmdjdias https://github.com/pmdjdias,

Check the results folder: [image: image] https://user-images.githubusercontent.com/73063947/160248254-f312feb2-5f49-4ce0-8099-552180f9365e.png

The folder contains the following information:

• Predicted and real pose for each frame
• csv file with the position and rotation error for each frame, and the average values
• config file with information regarding the dataset and model used.

— Reply to this email directly, view it on GitHub https://github.com/DanielCoelho112/localization_end_to_end/issues/23#issuecomment-1079726687, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACWTHVUTBSWIAEGOVJZKLQLVB424FANCNFSM5RIY3ANA . You are receiving this because you were mentioned.Message ID: @.***>

[DanielCoelho112]

Hi @miguelriemoliveira,

The csv file should have the units.

Yes, you're right.

Is it meters? If so that's a lot right?

This model was trained on my computer, it took 20 seconds to train. And it only used 2000 points and 90 frames... This is just to test the pipeline. Given the training conditions, I think the results were surprisingly good...

[miguelriemoliveira]

ok, great. 1 meter accuracy for 20 secs training is not bad...

[DanielCoelho112]

Done.