Question about dataset_convert tool

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

hello

I have two dataset

  1. dataset for training (18000 image & labels)
  2. dataset for valid ( 3000 images & labels)

According to the DetectNet_v2 — Transfer Learning Toolkit 3.0 documentation doc

It appears to separate one dataset into training and valid.

kitti_config {
  root_directory_path: "/workspace/tlt-experiments/data/VOCtrainval_11-May-2012/VOCdevkit/VOC2012"
  image_dir_name: "JPEGImages_kitti/test"
  label_dir_name: "Annotations_kitti/test"
  image_extension: ".jpg"
  partition_mode: "random"
  num_partitions: 2
  val_split: 20
  num_shards: 10
}
image_directory_path: "/workspace/tlt-experiments/data/VOCtrainval_11-May-2012/VOCdevkit/VOC2012"

but i want to make training set using my ‘dataset for training’ and make valid set using my ‘dataset for valid’.

this is possible??

3

It is possible.
Please generate tfrecords separately for your training dataset and validation dataset.

Then in training spec as below, set the tfreocrds files of your validation dataset.

validation_data_source: {
tfrecords_path: " /tfrecords validation pattern>"
image_directory_path: " "
}

and comment out

#validation_fold: 0

More info, see DetectNet_v2 — Transfer Learning Toolkit 3.0 documentation

4

i successed to transfer learning based on peopleNet

thanks to your advise Morganh