How to get all the parameter of labels file using labelImg for Training on TLT

Hello morganh,

Actually I have looked into http://www.cvlibs.net/download.php?file=data_object_image_2.zip dataset and their label file http://www.cvlibs.net/download.php?file=data_object_label_2.zip
which were mention in TLT. Their in the label file I am getting
Car 0.00 0 -1.67 657.39 190.13 700.07 223.39 1.41 1.58 4.36 3.18 2.27 34.38 -1.58
but when I do annotation on my image using LabelImg and convert into kitti then i got
Car 0.00 0 0.0 657.39 190.13 700.07 223.39 0.00 0.00 0.00 0.00 0.00 0.00 0.00

so how should I go so that I can get non zero value at position {1,2,3,8,9,10,11,12,13,14} , which annotation tool will give non zero value. please help.

and second is :
This DontCare is using as a negative data ?
DontCare -1 -1 -10 513.37 159.27 558.64 198.01 -1 -1 -1 -1000 -1000 -1000 -10
if then how we can add DontCare in our detectnet_v2_train_resnet18_kitti.txt file to make more accurate our model. we should treat DontCare as a class or something else.
Thanks

Hi solanki,
See the user guide.
Currently, for detection the TLT only requires the class name and bbox coordinates fields to be populated. This is because the TLT training pipe supports training only for class and bbox coordinates. The remaining fields maybe set to 0. Here is a sample file for a custom annotated dataset:

car 0.00 0 0.00 587.01 173.33 614.12 200.12 0.00 0.00 0.00 0.00 0.00 0.00 0.00

thanks for your reply.
please let me know about my second question
This DontCare is using as a negative data ?
DontCare -1 -1 -10 513.37 159.27 558.64 198.01 -1 -1 -1 -1000 -1000 -1000 -10
if then how we can add DontCare in our detectnet_v2_train_resnet18_kitti.txt file to make more accurate our model. we should treat DontCare as a class or something else.

Thanks.

You can trigger jupyter notebook inside the TLT docker to have a try.
We did not train “Don’t care” class.

Use the examples : Examples using ResNet18 backbone for detecting objects with either DetectNet_v2, SSD, or FasterRCNN architectures are available as Jupyter Notebooks. To run the examples that are available, enable the jupyter notebook included in the docker to run in your browser:

docker run --runtime=nvidia -it -v /home//tlt-experiments:/workspace/tlt-experiments -p 8888:8888 tlt-streamanalytics:

Go to the examples folder: cd examples/

Execute this command from inside the docker to start the jupyter notebook:

jupyter notebook --ip 0.0.0.0 --allow-root