Pre-collected data and use it on jetson-inference

[Hamzeh.nv]

Hi
I have collected the dataset previously and the label format of images is in Yolo format
and now I want to use them in jetson-inference docker and train ssd-mobile net
how can I do this?
@dusty_nv

[dusty_nv]

Hi @Hamzeh.nv, you might want to check if CVAT tool can convert them to Pascal VOC format for you.

Or here is a Python script I just found (but haven’t tried) that converts from YOLO to VOC format: https://github.com/carolinepacheco/Convert-YOLO-to-PascalVOC

[Hamzeh.nv]

@dusty_nv
what if I want to train the SSD model in another system and then Inference in Jetson?
I convert the labels and started training, but it seems to have a problem with annotations…
I want to train and convert the model to onnx in another system and then infer in jetson with jetson-inference docker container.

[dusty_nv]

@Hamzeh.nv

what if I want to train the SSD model in another system and then Inference in Jetson?

@Hamzeh.nv yes, you can clone the https://github.com/dusty-nv/pytorch-ssd directly to another machine and run the SSD training there, export it to ONNX, and then copy the ONNX model to your Jetson.

Your PC that runs the training should have an NVIDIA GPU, and I have run it with Ubuntu x86_64 with PyTorch installed - on x86, typically I run it inside the NVIDIA pytorch container from NGC (https://catalog.ngc.nvidia.com/orgs/nvidia/containers/pytorch)