Jetson-inference training Question

[apexis87]

Hallo,

I am not a professional, doing it for fun and learn something new. Up till now following the guides i have managed to get object detection up and running no problem.
Also using the GUI tool for camera capture.

I am wondering is there a way to use an existing dataset of images that i have … but i have no annotations.
How do i train the model? Any suggestions? When i use the python3 train_ssd.py … i get the no annotations .

thanx
-N

[dusty_nv]

@apexis87

I am wondering is there a way to use an existing dataset of images that i have … but i have no annotations.

Hi @apexis87, to annotate data that you have already captured, I use the CVAT tool for that:

• https://cvat.org/
• GitHub - opencv/cvat: Annotate better with CVAT, the industry-leading data engine for machine learning. Used and trusted by teams at any scale, for data of any scale.

You can run it from your web browser without needing to install anything. Then have CVAT export your annotated dataset in Pascal VOC format. After you copy the annotated dataset to your Jetson and extract it, add a labels.txt file to it (CVAT doesn’t create this).

Then you can run train_ssd.py like you did after you used the camera-capture tool:

• https://github.com/dusty-nv/jetson-inference/blob/master/docs/pytorch-collect-detection.md#training-your-model

Since both camera-capture tool and the CVAT tool export the data in Pascal VOC format, they both work the same way in train_ssd.py

[apexis87]

thank you