Train YOLOv10 on Custom Dataset with Ultralytics

Поділитися
Вставка
  • Опубліковано 20 гру 2024

КОМЕНТАРІ • 26

  • @NicolaiAI
    @NicolaiAI  5 місяців тому +3

    Join My AI Career Program
    www.nicolai-nielsen.com/aicareer
    Enroll in the Investing Course outside the AI career program
    nicolai-nielsen-s-school.teachable.com/p/investment-course
    Camera Calibration Software and High Precision Calibration Boards
    camera-calibrator.com/

    • @dakaradakaradakara699
      @dakaradakaradakara699 5 місяців тому

      Sir, Please make tutorial for object detection with image classier on real time

    • @TheidiotAmongUs
      @TheidiotAmongUs 5 місяців тому

      @@dakaradakaradakara699 if you mean a real time object detection video on webcam he did make one for yolov8 a while back

  • @sagnumb
    @sagnumb 5 місяців тому +3

    Just saw some reddit post about ultralytics yolov10, I'm not going to train my custom datasets using their wrapper due to the shady license.

    • @NicolaiAI
      @NicolaiAI  5 місяців тому

      Hmm what did u read? Don’t think there is any shady stuff, it’s just an open source license with no commercial use without a license or not making the code available. Good for many things and at the end of the day someone has to pay the bills. Atleast they put stuff out there to help push the space. The big guys like Google, meta etc have tons of open source models but they are not well supported and outdated

    • @sagnumb
      @sagnumb 5 місяців тому

      @@NicolaiAI Exactly why I wouldn't use their code to train my custom model, because my purpose is being able to commercialize my app one day. and me being someone that has low budget and trying to start my own business, I don't have money to pay for their license, I would prefer to use a fully open source model and not get sued at the end of the day, I thought yolo was suppose to be open source, but unfortunately Ultralytics has got a hold of most yolo nowadays.

    • @NicolaiAI
      @NicolaiAI  5 місяців тому

      @sagnumb yeah pretty much all models are build on top of Ultralytics. Good benchmark against other open source models. Check out the rtdetr

  • @shashankgsharma0901
    @shashankgsharma0901 5 місяців тому +1

    can you do fintuning of LLMs also like LLaVA-NeXT? It would be great!

    • @NicolaiAI
      @NicolaiAI  5 місяців тому

      Will definitely look into that! Thanks

  • @ignis.valorant
    @ignis.valorant 5 місяців тому +1

    Nice video as always, super helpful! I want to ask whether it is possible to use SAHI as a part of the training instead of the final inferences. My objects are kinda small, so I assume that using SAHI in the training phase can help a lot. I tried using tiles but it ended up being too slow to train unfortunately. Any advice is greatly appreciated

    • @NicolaiAI
      @NicolaiAI  5 місяців тому +1

      Hmm yeah with roboflow you can add tiling windows as a preprocessing stop for your dataset or do it manually. It will multiple your dataset size with the number of tiles so training will take that longer

    • @ignis.valorant
      @ignis.valorant 5 місяців тому

      ​@@NicolaiAI thanks for the reply! So is it safe to assume that SAHI is only used when making the final inference? For example, you train an object detection model for cars seen from above (satellite data). It can make accurate inferences when the image is clear. But suppose the cars seen in another satellite image are smaller and undetectable, SAHI can help with that?

  • @UpadhyaX
    @UpadhyaX 5 місяців тому +1

    hi sir I am from India I have been studying AIML engineering for 2 years 3 Sem at Mysore University can you make a video about the ai class about beginner to Hero level class for students to explain AI and how to code for AI please make a video AI notes with teaching about AI for beginner thank you, sir, making a video about ai

  • @RJR.Aca-ac
    @RJR.Aca-ac 4 місяці тому +1

    How to hyperparameter tune in training YOLOv10 models?

    • @NicolaiAI
      @NicolaiAI  4 місяці тому

      U can specify all the arguments in the train function. All the arguments can be found on the ultralytics docs

    • @RJR.Aca-ac
      @RJR.Aca-ac 4 місяці тому

      @@NicolaiAI Oh, thanks a lot for the response! Could you maybe share about choosing the right hyperparameters? Your help would mean a lot!

  • @SherinRS-k2i
    @SherinRS-k2i 5 місяців тому +1

    Hey Nicolai just want to tell you that i have just started learning Python. If i buy your course of AI career program will i be able to reach your level.

    • @NicolaiAI
      @NicolaiAI  5 місяців тому +1

      Hey awesome man! Cool that you are getting started on coding and ai. Yeah for sure, all my technical courses are in there as well together with the AI career stuff. We have tons of cool guys in there as well in the community and I help each one of the members with their own journey based on where they want to be and their current situation. But it will require hard work!

  • @fajrihanggi9759
    @fajrihanggi9759 4 місяці тому

    i alrd follow step by step but this eror said
    ScannerError: while scanning a simple key
    in "", line 11, column 1:
    test:/content/dsfs-8/test/images
    ^
    could not find expected ':'
    in "", line 12, column 1:
    train:/content/dsfs-8/train/images
    i dont even know where is line 12 and how i check it

  • @SlfgjkAldfjgf
    @SlfgjkAldfjgf 4 місяці тому

    How to set the augmentation paramters?
    # Load the pretrained YOLOv8 model
    model = YOLO(model_path) # Use the .pt file for pretrained weights
    # Define custom augmentation parameters
    augmentation_params = {
    'degrees': 19, # Rotation in degrees
    'translate': 0.1, # Translation as a fraction of image size
    'scale': 0.5, # Scaling factor
    'shear': 2.0, # Shear angle in degrees
    'perspective': 0.0, # Perspective transformation
    'flipud': 0.5, # Vertical flip probability
    'fliplr': 0.5, # Horizontal flip probability
    'mosaic': 1.0, # Mosaic augmentation probability
    'mixup': 0.2, # Mixup augmentation probability
    'hsv_h': 0.015, # HSV hue augmentation
    'hsv_s': 0.7, # HSV saturation augmentation
    'hsv_v': 0.4, # HSV value augmentation
    }
    # Additional training parameters
    training_params = {
    'data': dataset_path, # Path to your dataset
    'imgsz': 640, # Image size
    'augment': True, # Enable augmentation
    'patience': 10, # Early stopping patience
    'save_period': 1, # Save the model after every epoch
    'save': True, # Enable saving of model checkpoints
    'resume': True, # Resume training from the last checkpoint
    'project': save_dir, # Directory to save the project
    **augmentation_params # Include custom augmentation parameters
    }
    # Train the model with custom parameters
    model.train(**training_params)
    This didn't work.

  • @Ultralytics
    @Ultralytics 5 місяців тому +1

    Excellent overview of YOLOv10 with Ultralytics!

  • @arisalikbaev876
    @arisalikbaev876 5 місяців тому

    Good time of the day! Has anyone encountered this issue with DetMetrics curves TODO while training your model on yolov10:
    AttributeError: 'DetMetrics' object has no attribute 'curves_results'. See valid attributes below.
    How to deal with that?