YOLO-NAS vs YOLOV8 for Real-time Object Detection - Pros and Cons

Поделиться
HTML-код
  • Опубликовано: 22 дек 2024

Комментарии • 35

  • @NicolaiAI
    @NicolaiAI  Год назад +1

    Join My AI Career Program
    www.nicolai-nielsen.com/aicareer
    Enroll in My School and Technical Courses
    www.nicos-school.com

  • @FatemehZaremehrjardi
    @FatemehZaremehrjardi Год назад +2

    Thank you for the video! looking forward to watch your video about how to improve YoloNAS inference speed!

    • @NicolaiAI
      @NicolaiAI  Год назад

      Thanks a lot for watching! Have u been trying it out urself?

    • @FatemehZaremehrjardi
      @FatemehZaremehrjardi Год назад +1

      @@NicolaiAI I tried YoloNAS on my webcam and had the same issue but I think I should try the INT8 quantized version next time.

    • @NicolaiAI
      @NicolaiAI  Год назад +1

      Yeah for sure. Should still run way faster doe

  • @helloansuman
    @helloansuman Год назад

    How to train this on 8 GPUs on DGX A100?

  • @ashberten6884
    @ashberten6884 Год назад +1

    Thank you for your awesome tutorial. Do these models (YOLOv8 & YOLO-NAS) also work on Android? Please share your experience if you have tried it to deploy on Android.

  • @polymir9053
    @polymir9053 8 месяцев назад +1

    The reason for the slow performance with Yolo-nas is because your running inference using their .predict function. Their prediction functions are slow and not meant to be ran in a production system. You have to export the model into into something like onnx format to get the actual speeds. Once you do that, it should be much faster.

    • @NicolaiAI
      @NicolaiAI  8 месяцев назад

      Yup I know, but even in onnx it gets smoked by yolov8

    • @polymir9053
      @polymir9053 8 месяцев назад

      @@NicolaiAI Thats interesting.. I'm starting to doubt their claims that its faster. But looking back at their github, they seem to claim only the 8bit quant models are faster, so maybe they aren't lying? I'll have to try the 8 bit

  • @sumanpahari4220
    @sumanpahari4220 Год назад +1

    Thank you so much for very good contain.Can we use any other pretrained model instead of coco for YOLO NAS.

  • @jonathanr4242
    @jonathanr4242 Год назад +1

    Thank you, for sharing, Nicolai.

    • @NicolaiAI
      @NicolaiAI  Год назад

      Thanks for watching Jonathan!

  • @ZeynepC3
    @ZeynepC3 10 месяцев назад +1

    Thanks for the video, I have a question. Which one is faster, yolov8 or yolo-NAS? Have you had a chance to test this?

    • @NicolaiAI
      @NicolaiAI  10 месяцев назад

      Yolov8 for sure!

  • @pixeloverflow
    @pixeloverflow Год назад +1

    Thanks for putting this together!

    • @NicolaiAI
      @NicolaiAI  Год назад +1

      Thanks a lot for watching! Means a ton to me

  • @sabrimas9718
    @sabrimas9718 Год назад

    i tried yolov8 for sometime, and when i want to try tuning hyperparameters using raytune, it shows an error, even though i followed the steps provided by Ultralytics, can you make a video about tuning yolov8 hyperparameters using raytune?

  • @rakeshkumarrout2629
    @rakeshkumarrout2629 Год назад

    Hey Nic. This video is amazing.i have a question for you. Can you help me giving the approach , how to build a helmet detection model for motorbikes and a seat belt detector for cars and other vehicles as a single project? Kindly take the pain to share the right approach and models to use for that.

  • @caiyu538
    @caiyu538 Год назад +1

    Great lectures.

    • @NicolaiAI
      @NicolaiAI  Год назад

      Thanks a lot man! Appreciate u watching

  • @NeuralNetwork-go5zn
    @NeuralNetwork-go5zn Год назад +1

    hello, really nice tutorial!
    I created a custom yolo detection with yolov8m as a base, but when I run it, as soon as the speed of the objects increases, the algorithm loses track of me and starts to "jerk" the video output. Does anyone have an idea how to fix this?
    (yolov8m algorithm trained on 300 custom images)
    Thanks so much to anyone who can help me!!

    • @NicolaiAI
      @NicolaiAI  Год назад

      Hi Thanks a lot man! Are u using a tracker?

    • @NeuralNetwork-go5zn
      @NeuralNetwork-go5zn Год назад

      @@NicolaiAI yes i think, i use model.track(..)

  • @nhatpham5797
    @nhatpham5797 Год назад +1

    Hi, what graphics card are you using?

  • @caiyu538
    @caiyu538 Год назад +2

    I did a comparison between yolo and detr in a limited datasets and found yolo is a little better even though both of them results are far from satisfaction. Our training datasets are only around 200.

    • @NicolaiAI
      @NicolaiAI  Год назад

      Cool comparison. I don’t think much has been done to detr recently

    • @Craskar
      @Craskar Год назад

      I'm actually surprised that YOLO only did a little better. While transformer models tend to outperform pure CNNs, they are also know to be very data hungry as well.

  • @Polygone75
    @Polygone75 Год назад +3

    Love your content but I'm surprised that you didn't use more advanced comparators in the pros and cons like licenses, label format to train on custom data, ... it could have been more informative IMHO

    • @NicolaiAI
      @NicolaiAI  Год назад +2

      Thanks a lot for watching a the feedback! This was just a quick overview and to more to see how you can run the two models easily. We are definitely going to cover way more of all of that in the future. Stay tuned!

    • @maxfowler8838
      @maxfowler8838 Год назад +1

      @@NicolaiAI 🤝

  • @greendsnow
    @greendsnow Год назад +1

    Personally, I would detect you in an instant if I saw you outside :D
    1ms, piew!
    [That Yolo Guy]

    • @NicolaiAI
      @NicolaiAI  Год назад +1

      Hahah! Come say hi if you detect me one day