Works well but needs improvement
merts1 opened this issue · comments
Like all the other models, it is probably trained on youtube edits and model sometimes can not differentiate between teammates and enemies. Also most of those youtube clips consist of kills where you face the enemy, that's why these models fail to recognize enemies when you're behind them.
I think there is no need for improvement on recognition part, however there is a need to add negative samples on training phase with teammates on the background.
I mean I would like to help this project too, I'm just writing this not as a criticisizm but as a way to encourage a better model.
Reall cool work!
edit: I have only tested [valorant-v6.pt], [valorant-n.pt] , [valorant-bot.pt] versions btw.
Actually, models are based on images from actual play, not YouTube.
And I'm already aware of the problem that almost models are unable to recognize between enemy and teammate. But I won't make any changes to valorant-v and valorant-n series.
I gonna make another version that can recognize. And I will be distributing a dataset that was used for valorant-v and valorant-n series.
Thanks for your support.
And I will be distributing a dataset that was used for valorant-v and valorant-n series.
That would be a great contribution.
https://mega.nz/file/XM4DgCSK#-esZ_rkH8val8SuB8nCiGuw4bpHITZVZjYaxYO6Tcb8
This is another data collection if you want to train on your model.
Thanks for sharing your data collection.
It may be used to train 😄
I’ve looked your dataset over.
It would be very helpful for training.
Thanks 😄
As you can see, there are two annotations in your shared file in clip2(1).txt.
I can understand one of the annotations that is "enemy", but I cannot understand the other.
What is it?
As you can see, there are two annotations in your shared file in clip2(1).txt. I can understand one of the annotations that is "enemy", but I cannot understand the other. What is it?
can see that by openin the label file with notepad
I see!
Thanks 👍🏻
I see! Thanks 👍🏻
you're welcome, I'm always open to help this project
I released a dataset.
Check it out.
I will try to create a color map ai detection method by combining these data, I think that would help the model.
It sounds great.
I'm looking forward to it!
I released a dataset. Check it out.
Hi, im wondering if you could post a dropbox/alternative download link for the dataset? I couldn't seem to be able to download them on github, as it says page not found. Thanks!