You Only Look Once: object detection with YOLOv4 | DIY drone pt. 3
ฝัง
- เผยแพร่เมื่อ 5 มิ.ย. 2024
- YOLO: ain't that from the 2010s? THINK AGAIN. Its all about forcing electrocuted sand to see what stuff is in photos and video, like cats and skateboards.
Boring? Don't be so fast to judge! You can do some REALLY cool stuff with a bit of YOLO magic.
Catch up with the rest of this series here:
Part 1: • My Raspberry Pi drone:...
Part 2: • Stereo depth mapping w...
Part 3: (this video)
Part 4: • Meet Stanley, my NVIDI...
Part 5: • Teaching my custom AI ...
In this video, I'm covering:
- What the heck is YOLO?
- History of YOLO
- Performance tests
00:00 - Intro
00:38 - The release of YOLO
01:26 - Inject more YOLO into my veins
01:55 - Into the multiverse
03:01 - Outdoor distance tests
04:59 - Inference speed on different hardware
05:49 - Speed up by scaling down
06:18 - Wrapping up
🌐 Links
YOLO
pjreddie.com/darknet/yolo/
YOLO vs R-CNN
towardsdatascience.com/r-cnn-...
Controversy over YOLOv5
blog.roboflow.com/yolov4-vers...
Benchmarks
/ yolov4-vs-yolov4-tiny
/ yolo3-a-huge-improvement
/ yolov4-the-most-accura...
Timeline
towardsdatascience.com/yolo-v...
Repositories
YOLO thru YOLOv3: github.com/pjreddie/darknet
YOLOv4: github.com/AlexeyAB/darknet
YOLOv5: github.com/ultralytics/yolov5
PP-YOLO (1 and 2): github.com/PaddlePaddle/Paddl...
Scaled-YOLOv4: github.com/WongKinYiu/ScaledY...
YOLOR: github.com/WongKinYiu/yolor
ℹ️ Attributions
2080Ti Footage (JayzTwoCents)
• RTX 2080 and 2080Ti Be...
🎵 Music
bezserca - purpose ( • bezserca - purpose )
Vans in Japan - Deep State ( • Video )
Punch Deck - Rise ( / rise )
💡 Research Papers
You Only Look Once: Unified, Real-Time Object Detection. Redmon, J. et al. 2016.
arxiv.org/abs/1506.02640
YOLOv4: Optimal Speed and Accuracy of Object Detection. Bochkovskiy, A. et al. 2020.
arxiv.org/abs/2004.10934
Scaled-YOLOv4: Scaling Cross Stage Partial Network. Wang, C-Y. et al. 2021
arxiv.org/abs/2011.08036
🧐 Social stuff
Twitter: / akamatchstic - วิทยาศาสตร์และเทคโนโลยี
Like my favorite t-shirt says. "I don't know how to act my age, I have never been this old before" Love Your Work
I commented on your other video.3 years ago I used this ML algo to determine states on a video, had to train a model myself for it long story short with little detail there. This absolutely is one of the best techniques for quick OCR image detection.
Please keep going with this as your starting knowledge is ahead of mine you’ll give me a head start when I try to do this.
Superb video and mind blowing cool stuff!!!
Very cool! Had a play with YOLO for a uni project last year, and it really is amazing
Thanks! Yep it sure is a fantastic piece of research, especially now that it’s effectively supported by a wider community of researchers
Your videos are so fun and interesting. Thank you.
Amazing, which dataset you use?
Hey, I am also working on the implementation of a Jetson Nano in my drone. Even working at 5W mode, however, I couldn't yet find an appropriate way to power the Jetson at a stable voltage. Linked to two cameras and writing the (still) captures constantly in the memory, the voltage of two parallel 5A buck converters (over the jack plug) still drops at ~4.8V or sometimes even below 4.6V.
How do you solve that issue?
Heyy I am also working on a similar project can you suggest which camera would be good to use with rpi4 for object detection tasks from a drone?
I'm curious, are you doing anything to prevent it following other people? Loving the knowledge you're dropping on us. Learning a lot in areas I wouldn't normally get exposure to.
To start with, probably not 😅 it’ll have to be a careful game of “follow the person that is closest” during testing, which is likely to be good enough
Why while watching this video, The Netflix's "YOU" tv series keep popping up on my mind?
what a project! i love to see some drone diy contents! since a lot of drone from DJI or any others, people starting leave a drone project. I'm interested on your videos, if u would notice my comment, it would be a really amazing video if u want to make some tutorials about making diy drone, from jetson nano too! because jetson really powerful than raspberry for AI impelementation. Cheers!
Thanks! I can definitely take a look into doing a tutorial or something. My current focus is on monthly projects with this drone being something worked on in the background.
Would you be ditching the heatsink on the nano? And most importantly, how would you be mounting it to the Martian? No space!
For now I think I’ll be keeping the current heatsink. As for mounting, that’s one part I haven’t shown properly yet! I’ve designed a mounting system on the underside of the drone, where I can attach hardware. Right now, I have another Pi Zero there which is how I’m getting aerial footage, which will be swapped for the Nano.
@@akamatchstic Understandable. The current flight stack serves the purpose nicely and the drone could certainly use some CUDA to YOLO your frames. Waiting for the next video to see if you've upgraded the prop size to something in the 10 inch neighborhood.
Where is the first part?
Here's a link: th-cam.com/video/ZCOlT_sz6Gs/w-d-xo.html