How to data Augmentation in classification training

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Please provide the following information when requesting support.

• Hardware (T4/V100/Xavier/Nano/etc)
Ubuntu 20.04.3 LTS x64 , RTX3090
• Network Type (Detectnet_v2/Faster_rcnn/Yolo_v4/LPRnet/Mask_rcnn/Classification/etc)
Classification
• TLT Version (Please run “tlt info --verbose” and share “docker_tag” here)

!tao info
dockers: [‘nvidia/tao/tao-toolkit-tf’, ‘nvidia/tao/tao-toolkit-pyt’, ‘nvidia/tao/tao-toolkit-lm’]
format_version: 2.0
toolkit_version: 3.21.11
published_date: 11/08/2021

• Training spec file(If have, please share here)

model_config {
  arch: "resnet",
  n_layers: 50
  # Setting these parameters to true to match the template downloaded from NGC.
  use_batch_norm: true
  all_projections: true
  freeze_blocks: 0
  freeze_blocks: 1
  input_image_size: "3,224,224"
}
train_config {
  train_dataset_path: "/workspace/tao-experiments/data/split/train"
  val_dataset_path: "/workspace/tao-experiments/data/split/val"
  pretrained_model_path: "/workspace/tao-experiments/classification/pretrained_resnet50/pretrained_classification_vresnet50/resnet_50.hdf5"
  optimizer {
    sgd {
    lr: 0.015
    decay: 0.0
    momentum: 0.9
    nesterov: False
  }
}
  batch_size_per_gpu: 48
  n_epochs: 120
  n_workers: 16
  preprocess_mode: "caffe"
  enable_random_crop: True
  enable_center_crop: True
  label_smoothing: 0.0
  mixup_alpha: 0.1
  # regularizer
  reg_config {
    type: "L2"
    scope: "Conv2D,Dense"
    weight_decay: 0.00005
  }

  # learning_rate
  lr_config {
    step {
      learning_rate: 0.009
      step_size: 10
      gamma: 0.1
    }
  }
}
eval_config {
  eval_dataset_path: "/workspace/tao-experiments/data/split/test"
  model_path: "/workspace/tao-experiments/classification/output/weights/resnet_120.tlt"
  top_k: 3
  batch_size: 256
  n_workers: 8
  enable_center_crop: True
}

• How to reproduce the issue ? (This is for errors. Please share the command line and the detailed log here.)
I noticed the accurate is surprisingly low when I test the sample image with upside down or other angle against the final trained model, but my training dataset does not cover those cases.
Question: Is there anyway in Training spec that can enable these image angle Augmentation?

3

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

You can use a tool mentioned in Offline Data Augmentation - NVIDIA Docs to generate more augmented training images offline. Then use them in the new training.

5

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.