TAO tao detectnet_v2 dataset_convert not converting KITTI file to tfrecords even if the SPECS file are correct

[bijay1]

Please provide the following information when requesting support.

• Hardware ntel® Core™ i5-10500H CPU @ 2.50GHz × 12 NVIDIA GeForce RTX 3060 Laptop GPU/PCIe/SSE2 (USING TAO docker)
• Network Type : Detectnet_v2
• TAO Version 3.22.02
• Training spec file Kitti file :

kitti_config {
root_directory_path: “/tao/data/training”
image_dir_name: “image”
label_dir_name: “label”
image_extension: “.png”
partition_mode: “random”
num_partitions: 2
val_split: 14
num_shards: 10
}
image_directory_path: “/tao”

• How to reproduce the issue ?

I have used the following command to convert the kitti dataset to tfrecords:

tao detectnet_v2 dataset_convert
-d $SPECS_DIR/train_kitti_config.txt
-o $LOCAL_DATA_DIR/tfrecords

Got the following output:

tao detectnet_v2 dataset_convert
-d $SPECS_DIR/train_kitti_config.txt
-o $LOCAL_DATA_DIR/tfrecords

And this is the result I am getting :

I don’t understand where I went wrong.
000000.txt (670 Bytes)

person 0.00 0 0.00 266 525 587 1048 0.00 0.00 0.00 0.00 0.00 0.00 0.00
person 0.00 0 0.00 1232 462 1514 1023 0.00 0.00 0.00 0.00 0.00 0.00 0.00

That is how my label files looks like.

Please point me to right direction.

[Morganh]

You have only 10 images. It is not enough.

Please set
num_shards: 10

to

num_shards: 1

and retry.

[bijay1]

Hi Morganh,

I tried that it didn’t work. I have also used the dataset provided in kitti format which has 7481 images with no result.

Here is the command
!tao detectnet_v2 dataset_convert
-d $SPECS_DIR/train_kitti_config.txt
-o $LOCAL_PROJECT_DIR/tfrecords

and config file:
kitti_config {
root_directory_path: “/tao/data/training”
image_dir_name: “image_2”
label_dir_name: “label_2”
image_extension: “.png”
partition_mode: “random”
num_partitions: 2
val_split: 14
num_shards: 10
}
image_directory_path: “/tao/data/training”

and this is the output but no files created on the directory.

[Morganh]

Could you please make sure if the result folder has written access?

[bijay1]

Thank you for the reply but the directory has read write permission

[Morganh]

To debug, please run below in terminal instead of notebook. I am afraid you need to check if ~/.tao_mounts.json is correct.

$ tao detectnet_v2

then

# detectnet_v2 dataset_convert xxx

[bijay1]

Result after running detectnet dataset_convert XXX in docker and ~/.tao_mounts.json looks fine

[bijay1]

Thank you. It worked