How the fine-tuning works for fasterRCNN? I need to add new lables to an *.tlt model

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Please provide the following information when requesting support.

• Hardware (T4/V100/Xavier/Nano/etc)
Any device
• Network Type (Detectnet_v2/Faster_rcnn/Yolo_v4/LPRnet/Mask_rcnn/Classification/etc)
Faster_rcnn
• TLT Version (Please run “tlt info --verbose” and share “docker_tag” here)
• Training spec file(If have, please share here)
• How to reproduce the issue ? (This is for errors. Please share the command line and the detailed log here.)

Considering that I have a trained model or I am using a pre-trained model with some objects and their associated labels (car and pedestrian). Assuming that I am receiving a new dataset and the images are just labels as trucks. on my training config file I had classes as follows:

target_class_mapping {
key: 'car'
value: 'car'
}
target_class_mapping {
key: 'pedestrian'
value: 'person'
}

Now I want to add the trucks to the model and fine-tune the last layers of the model. Now the questions are :
1- Should I use tao train or tao retain command?
2- Should I change any part of the config file? and use
freeze_bn: True
freeze_blocks: 0
freeze_blocks: 1
2- Should I keep the previous target classes in the config file and add the new labels as follows or I should remove them?

target_class_mapping {
key: 'car'
value: 'car'
}
target_class_mapping {
key: 'pedestrian'
value: 'person'
}
+
target_class_mapping {
key: 'trucks'
value: 'trucks'
}

Regards

3

You can

  1. use tao train
  2. set pretrained_weights to your previous trained model.
  3. keep the previous target classes in the config file and add the new labels
1 Like
5

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