• State TI Thinks Resolved
  • Locked Locked
  • Replies 3 replies
  • Answers 1 answer
  • Subscribers 98 subscribers
  • Views 87 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

EDGE-AI-STUDIO: Data augmentation in edgeai modelmaker

Sajan Saji
Intellectual 490 points
Part Number: EDGE-AI-STUDIO

Tool/software:

Hello Engineer,

Edgeai model maker supports data augmentation? By checking the generated models param.yaml, saw shuffle is true for training. Is it support transforms like changing image angle, crop image, adjust various brightness, contrast etc.

Warm Regards,
Sajan

  • 0
    TI__Genius 15156 points

    Hello Sajan,

    No, dataset augmentations are not explicitly supported by the tooling. Some models like YOLOX may use training frameworks that implicitly perform augmentations like stitching or scaling images. Transforms like brightness, cropping, contrast, noise, and other effects are not support in Edge AI Studio at this time. You would need to perform these yourself outside the tool and upload those images back into Edge AI Studio.

  • 0 in reply to Reese Grimsley
    Intellectual 490 points

    Hello Reese,

    Reese Grimsley said:
    Transforms like brightness, cropping, contrast, noise, and other effects are not support in Edge AI Studio at this time

    I have a doubt it your reply. Did you mean data augmentation in the training or the edgeai interface.

    This like code exists in edgeai modelmaker ?

    transforms.Compose([
    transforms.RandomResizedCrop(224),
    transforms.RandomHorizontalFlip(),
    transforms.ColorJitter(brightness=0.2, contrast=0.2, saturation=0.2, hue=0.1),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])

  • 0 in reply to Sajan Saji
    TI__Genius 15156 points

    Hello Sajan,

    edgeai-modelmaker does not expose an interface at the user-level for this. The code you are pointing to in edgeai-tensorlab is not part of edgeai-modelmaker, which is the user interface for the training process as supported by TI.

    For example, search results [0] show code that you include, but this is part of other open source projects that are incorporated into edgeai-tensorlab, and are not part of TI's interface into this tooling. Edgeai-modelmaker does not explicitly support this, but if you would like to modify the source code to include more augmentations, you are free to do so as with any open-source project. 

    https://github.com/search?q=repo%3ATexasInstruments%2Fedgeai-tensorlab%20%20transforms.RandomHorizontalFlip()&type=code