You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I know of a complaint for motorcycle plates, which have the same form factor as the trucks plates.
tfaehse said he doesn't have enough training data of them.
On the edges of the frame, the plates are bigger and thus need a larger --blur_size than the default. Especially with 4K footage, the blurring is simply not strong enough otherwise.
The edges still appear a bit sharp, so I would increase the --feather_edges option.
If it's possible for (either of) you, could you share some unedited footage that contains these license plates that aren't getting detected? I could try and manually label some of it and include it in the training.
Hello
I spend a bit time today to check the diffrent settings in the GUI Version with Yolo V8 (August 2023)
I used a 1min 4K Clip driving on a highway.
3840X2160p /H264-MPEG4 AVC/29.97fps/439MB
Also one Test with 1080p vs 4K
My Computer:
OS: Win11 (Latest)
CPU: i7 13700k
GPU: Nvidia RTX4090 (24GB)
RAM: 32GB DDR5 6000MHz
Storage: 2TB NMME
Only Issue I have are the Trucks licence plate not getting detected well from Switzerland:
See Video:
(url)](https://youtu.be/dOnSkP5UEac)
Any other advice I should try?
I`m happy with the result 1min 4K video takes 05:12min and 1080p version 01:42min
The text was updated successfully, but these errors were encountered: