So now, this is where you "recruit" (force) people to take pictures and then label said pictures, and "recruit" (force) other people to sift through the images and do quality control
Generation of labeled images could most likely be automated with SQF. There's a command for taking screenshots and the labeling could be derived from the screen positions of the bounding box .
Yes! A friend of mine saw my project and made a somewhat functional script that randomly places objects in a scene, takes pictures of them, and puts the bounding boxes. If we ever finish working on this script, I definitely had in mind to make another video on it, but I can't promise anything.
Great job! I really respect what you did. I have zero knowledge about programming, but I really love it. If you need an ARMA Pilot in the future, I will be happy to help.
Yeah pretty much. The dataset needs a good variety of backgrounds (weather, location, time) and objects (other vehicles, buildings, nature) in the military vehicles' pictures to better learn how to differentiate the military vehicle from everything else
You probably didn’t know but in Ukraine we do literally same thing but in real life. We have a lot of videos from FPV drones to train model detect enemies
Good to hear that this technology is starting to get practical. I saw articles on drones using YOLO models to do reconnaissance, but they were primarily for detecting vehicles, and the articles said the performance was still mediocre at this stage. What's your firsthand account of this? How is the performance?
I'm just curious about one thing though. The detection of CSAT vehicles we're seeing here, is it done live ingame using some kind of mod or injector or did you record the gameplay footage first then run yolo on it ?
This was recorded before hand and I then ran yolo on the footage. Yolo does not take a long time to run though so it would not be absurd to run a detection model like this in real time
Thanks! I don't know much about YOLO versions, I just asked my colleagues what version is stable and fitting for my use case, and they recommended v5 then.
Hey! I don't personally want to do it because I don't have cats in my area. If you want to work on this, it won't be difficult if you have some software engineering skills: YOLOv5 can already detect cats in images, and there are many tutorials on using YOLO on live cameras!
Thanks! Yes, if you're familiar with Python, the model and usage instructions are available here: github.com/AlexandreSajus/Military-Vehicles-Image-Recognition
@@Film-maker This requires some experience in python development. My project is a fork of YOLOv5. I suggest you look up tutorials online on how to use YOLOv5 on custom datasets. There is a lot of resources on the topic but it's not an easy task. Let me know if you get stuck on a specific step
@@alexandresajus Well im a gamer and im trying to do this with ark survival evolved, its also open world. I just started using python and chatgpt as my coder, only this chatgpt can't help much so if you can do a step by step tutorial it would be great! I get a lot of errors that i dont know to fix....
It might be tricky without some software engineering background. You can try looking at tutorials about YOLO Image Segmentation on Custom Datasets. Unfortunately, ChatGPT can only take you so far; learning Python and using libraries might take some time. I'll try to make a tutorial.
So now, this is where you "recruit" (force) people to take pictures and then label said pictures, and "recruit" (force) other people to sift through the images and do quality control
XDDDDDD
Generation of labeled images could most likely be automated with SQF. There's a command for taking screenshots and the labeling could be derived from the screen positions of the bounding box .
Yes! A friend of mine saw my project and made a somewhat functional script that randomly places objects in a scene, takes pictures of them, and puts the bounding boxes. If we ever finish working on this script, I definitely had in mind to make another video on it, but I can't promise anything.
Great job! I really respect what you did.
I have zero knowledge about programming, but I really love it.
If you need an ARMA Pilot in the future, I will be happy to help.
Thanks a lot! Yeah sure, if we see each other on a server: "You can be my wingman anytime!"
I got this idea but wasn't able to do it due to my lack of AI knowledge. Very cool video!
Thanks!
Yo! Great job working for something similat but for helicopters 😅
Nice! Have you shared your project anywhere so I could take a look?
So basicly: the larger the dataset of what IS NOT a military vehicle, the easier the NN can identifiy what IS an actual military vehicle?
Yeah pretty much. The dataset needs a good variety of backgrounds (weather, location, time) and objects (other vehicles, buildings, nature) in the military vehicles' pictures to better learn how to differentiate the military vehicle from everything else
You probably didn’t know but in Ukraine we do literally same thing but in real life. We have a lot of videos from FPV drones to train model detect enemies
Good to hear that this technology is starting to get practical. I saw articles on drones using YOLO models to do reconnaissance, but they were primarily for detecting vehicles, and the articles said the performance was still mediocre at this stage. What's your firsthand account of this? How is the performance?
I'm just curious about one thing though. The detection of CSAT vehicles we're seeing here, is it done live ingame using some kind of mod or injector or did you record the gameplay footage first then run yolo on it ?
This was recorded before hand and I then ran yolo on the footage. Yolo does not take a long time to run though so it would not be absurd to run a detection model like this in real time
Cool! Have you tried using it with videos from ongoing military conflicts or is it too rigid and tied to the game?
Unfortunately this would be too rigid, I would have to re-train it on real videos (a lot!) for it to work
Hey, Absolutely great project, Can you share the dataset too?
Sure! It's available here: www.kaggle.com/datasets/alexandresajus/arma3cvdataset
Thank you. very interesting. why did you choose v5 and not another version?
Thanks! I don't know much about YOLO versions, I just asked my colleagues what version is stable and fitting for my use case, and they recommended v5 then.
I am a vision engineer as well, if you need some help training the model on a type of vehicle let me know as I was also working on such a project
Thanks for the help. Unfortunately, I do not plan on continuing this project.
hey can you do this with cats? because I think knowing where all the cats in my neighborhood are would be a good thing to know
Hey! I don't personally want to do it because I don't have cats in my area. If you want to work on this, it won't be difficult if you have some software engineering skills: YOLOv5 can already detect cats in images, and there are many tutorials on using YOLO on live cameras!
@@alexandresajusthankyou for the advice wish me luck
Cool project. Is this model available?
Thanks! Yes, if you're familiar with Python, the model and usage instructions are available here: github.com/AlexandreSajus/Military-Vehicles-Image-Recognition
@@alexandresajus Would you be able to explain how I can do training with my images?
@@Film-maker This requires some experience in python development. My project is a fork of YOLOv5. I suggest you look up tutorials online on how to use YOLOv5 on custom datasets. There is a lot of resources on the topic but it's not an easy task. Let me know if you get stuck on a specific step
Can you do this again with more tutorial part? Would help me a lot!
I might do a tutorial on this. In the meantime, do not hesitate if you have any specific questions about the process.
@@alexandresajus Well im a gamer and im trying to do this with ark survival evolved, its also open world. I just started using python and chatgpt as my coder, only this chatgpt can't help much so if you can do a step by step tutorial it would be great! I get a lot of errors that i dont know to fix....
It might be tricky without some software engineering background. You can try looking at tutorials about YOLO Image Segmentation on Custom Datasets. Unfortunately, ChatGPT can only take you so far; learning Python and using libraries might take some time. I'll try to make a tutorial.
Thanks!@@alexandresajus
@@alexandresajusI'll be waiting your tutorial. It will be helpful for me
awesome! do you think this could ever make its way into a mod?
Could be possible, what do you have in mind?
Great... Now do this with NATO and RHS
Arghhh… *spends the rest of my life piloting drones and annotating images*
You're def getting banned from altis life.
Haha, you could probably make automated speed radars and surveillance cams for the altis police using this tech
Отлично. Молодец. С уважением из России
Спасибо!