TLT3.0 KITTI to TFRecords

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Hi, trying to convert custom KITTI to TFRecords.

/thx!

Creating a new directory for the output tfrecords dump.

print(“Converting Tfrecords for kitti trainval dataset”)
!detectnet_v2 dataset_convert
-d $SPECS_DIR/detectnet_v2_tfrecords_kitti_trainval.txt
-o $DATA_DOWNLOAD_DIR/tfrecords/kitti_trainval/kitti_trainval

Converting Tfrecords for kitti trainval dataset
Using TensorFlow backend.
WARNING:tensorflow:Deprecation warnings have been disabled. Set TF_ENABLE_DEPRECATION_WARNINGS=1 to re-enable them.
Using TensorFlow backend.
2021-02-14 15:51:23,889 - iva.detectnet_v2.dataio.build_converter - INFO - Instantiating a kitti converter
2021-02-14 15:51:23,894 - iva.detectnet_v2.dataio.kitti_converter_lib - INFO - Num images in
Train: 2102 Val: 342
2021-02-14 15:51:23,894 - iva.detectnet_v2.dataio.kitti_converter_lib - INFO - Validation data in partition 0. Hence, while choosing the validationset during training choose validation_fold 0.
2021-02-14 15:51:23,895 - iva.detectnet_v2.dataio.dataset_converter_lib - INFO - Writing partition 0, shard 0
WARNING:tensorflow:From /home/vpraveen/.cache/dazel/_dazel_vpraveen/216c8b41e526c3295d3b802489ac2034/execroot/ai_infra/bazel-out/k8-fastbuild/bin/magnet/packages/iva/build_wheel.runfiles/ai_infra/iva/detectnet_v2/dataio/dataset_converter_lib.py:142: The name tf.python_io.TFRecordWriter is deprecated. Please use tf.io.TFRecordWriter instead.

2021-02-14 15:51:23,895 - tensorflow - WARNING - From /home/vpraveen/.cache/dazel/_dazel_vpraveen/216c8b41e526c3295d3b802489ac2034/execroot/ai_infra/bazel-out/k8-fastbuild/bin/magnet/packages/iva/build_wheel.runfiles/ai_infra/iva/detectnet_v2/dataio/dataset_converter_lib.py:142: The name tf.python_io.TFRecordWriter is deprecated. Please use tf.io.TFRecordWriter instead.

/usr/local/lib/python3.6/dist-packages/iva/detectnet_v2/dataio/kitti_converter_lib.py:273: VisibleDeprecationWarning: Reading unicode strings without specifying the encoding argument is deprecated. Set the encoding, use None for the system default.
Traceback (most recent call last):
File “/home/vpraveen/.cache/dazel/_dazel_vpraveen/216c8b41e526c3295d3b802489ac2034/execroot/ai_infra/bazel-out/k8-fastbuild/bin/magnet/packages/iva/build_wheel.runfiles/ai_infra/iva/detectnet_v2/scripts/dataset_convert.py”, line 90, in
File “/home/vpraveen/.cache/dazel/_dazel_vpraveen/216c8b41e526c3295d3b802489ac2034/execroot/ai_infra/bazel-out/k8-fastbuild/bin/magnet/packages/iva/build_wheel.runfiles/ai_infra/iva/detectnet_v2/scripts/dataset_convert.py”, line 86, in main
File “/home/vpraveen/.cache/dazel/_dazel_vpraveen/216c8b41e526c3295d3b802489ac2034/execroot/ai_infra/bazel-out/k8-fastbuild/bin/magnet/packages/iva/build_wheel.runfiles/ai_infra/iva/detectnet_v2/dataio/dataset_converter_lib.py”, line 71, in convert
File “/home/vpraveen/.cache/dazel/_dazel_vpraveen/216c8b41e526c3295d3b802489ac2034/execroot/ai_infra/bazel-out/k8-fastbuild/bin/magnet/packages/iva/build_wheel.runfiles/ai_infra/iva/detectnet_v2/dataio/dataset_converter_lib.py”, line 105, in _write_partitions
File “/home/vpraveen/.cache/dazel/_dazel_vpraveen/216c8b41e526c3295d3b802489ac2034/execroot/ai_infra/bazel-out/k8-fastbuild/bin/magnet/packages/iva/build_wheel.runfiles/ai_infra/iva/detectnet_v2/dataio/dataset_converter_lib.py”, line 146, in _write_shard
File “/home/vpraveen/.cache/dazel/_dazel_vpraveen/216c8b41e526c3295d3b802489ac2034/execroot/ai_infra/bazel-out/k8-fastbuild/bin/magnet/packages/iva/build_wheel.runfiles/ai_infra/iva/detectnet_v2/dataio/kitti_converter_lib.py”, line 176, in _create_example_proto
File “/home/vpraveen/.cache/dazel/_dazel_vpraveen/216c8b41e526c3295d3b802489ac2034/execroot/ai_infra/bazel-out/k8-fastbuild/bin/magnet/packages/iva/build_wheel.runfiles/ai_infra/iva/detectnet_v2/dataio/kitti_converter_lib.py”, line 279, in _add_targets
AssertionError: Ground truth kitti labels should have only 15 fields.
Traceback (most recent call last):
File “/usr/local/bin/detectnet_v2”, line 8, in
sys.exit(main())
File “/home/vpraveen/.cache/dazel/_dazel_vpraveen/216c8b41e526c3295d3b802489ac2034/execroot/ai_infra/bazel-out/k8-fastbuild/bin/magnet/packages/iva/build_wheel.runfiles/ai_infra/iva/detectnet_v2/entrypoint/detectnet_v2.py”, line 12, in main
File “/home/vpraveen/.cache/dazel/_dazel_vpraveen/216c8b41e526c3295d3b802489ac2034/execroot/ai_infra/bazel-out/k8-fastbuild/bin/magnet/packages/iva/build_wheel.runfiles/ai_infra/iva/common/entrypoint/entrypoint.py”, line 296, in launch_job
AssertionError: Process run failed.

Also while trying to download the new sample notebooks while inlogged via NGC.

ngc registry resource download-version "nvidia/tlt_cv_samples:v1.0"

Generates:

load-version “nvidia/tlt_cv_samples:v1.0”
{
“error”: “Error: ‘nvidia/tlt_cv_samples:v1.0’ could not be found.”
}

2

For “AssertionError: Ground truth kitti labels should have only 15 fields.” , please refer to

For TLT 3.0, please wait for official annoucement.

3

Hi,

@Morganh
Thanks!

What is the recommended procedure to convert common formats as e.g Pascal VOC/YOLO to KITTI?

4

What we talking about “KITTI” here is KITTI format label file. See
https://docs.nvidia.com/metropolis/TLT/archive/tlt-20/tlt-user-guide/text/preparing_data_input.html#kitti-file-format

5

Try to run the command without specifying the version:
ngc registry resource download-version “nvidia/tlt_cv_samples”
This will output:
No version specified, downloading latest version: ‘v1.0.1’

1 Like
6

This is how a got it to work,

  1. Annotate in label tool of choice.

  2. Upload to Roboflow and download as YOLO Darknet.

  3. Convert YOLO to KITTI here → https://github.com/eweill/convert-datasets

  4. Convert in Detectnet_V2 notebook to TFrecords

All done!

7

I’ll check it out! Thanks

8

Hi @Morganh

We have our annotation in tfrecords format generated from RoboFlow framework for our custom dataset.
Can we use that tfrecords directaly in TLT?

9

Not sure if it can work, you can try.
If you already have the KITTI format labels, it is still recommended to use dataset_convert to generate the tfrecords.