Adding new object to ssd mobilev2 or other networks

how can i add a new objects to ssd-mobilenet v2. When install jetson-inference, there are 91 objects, how can i add a new objects,

Hi,

To add an new object to the model, please re-train the ssd-mobilenet-v2 with the new database.
A similar process can be found here:

Thanks.

when i want to run detectnet.py or my-detection.py or etc.in jetson-inference/build/aarch64/bin directory, there are 91 object to recognize for ssd-mobilenet-v2.

My question is how could i add a new object to know that found in jetson-inference/build/aarch64/bin directory?

Hi,

What kind of the new object do you want to add.
If the new object doesn’t exist in the 91 class, please retrain the model with an expanded database.

Thanks.

buoy, barge, canoe, paddle, jetski.
How retrain?

from here

I could not find ‘buoy’ in the Open Images dataset, so for that buoy class you may have to use a similar class, or find the data elsewhere (or collect your own).

However the other classes are in Open Images, so you could download them like so:

$ python3 open_images_downloader.py --class-names "barge,canoe,paddle,jetski" --data=data/my_dataset

Then you would proceed with the training as shown in the GitHub page.

Note that this would make a 4-class model, not added to the 91-class MS COCO model.

OK. when i run the detectnet.py in jetson-inference/build/aarch64/bin directory, it will recognize the new object OK thank you.

By the way could i use the marking objects in images like

https://github.com/tzutalin/labelImg or
https://github.com/opencv/cvat or etc.

İf can use how can i do this

labelImg and cvat should work - choose to export in Pascal VOC format.

1 Like

Thank you