None Value handling in multi-task classification

Dear @Morganh and Team,

I want to train a classification model to classify three items: Helmet, Vest, and Shoes within a single person bounding box. I do not want to crop the person into three separate classes.

How should I prepare the data to train the model with TAO?

For instance, if I have an image of a full person where shoes are missing, the output should be like (Helmet, Vest, None) or if class present then output should be (Helmet,Vest,Shoes).

When all three classes are present in a single person, it is easy to train a multi-task classification model. However, some instances may have a missing class with a value of None. Please suggest how to structure the dataset for this model, and which samples can help in training.

Thanks.

It can be also trained with multi-task classification.
Refer to the table in Multitask Image Classification - NVIDIA Docs.
You can set something like
image

Thanks @Morganh

Is there any sample application in deepstream for this type of multi-task classification (multi-task binary classification) model ? Where I can test this model easily.

Thanks.

It is available in the official deepstream_tao_apps. GitHub - NVIDIA-AI-IOT/deepstream_tao_apps: Sample apps to demonstrate how to deploy models trained with TAO on DeepStream.
The config files can be found in deepstream_tao_apps/configs/nvinfer/multi_task_tao at master · NVIDIA-AI-IOT/deepstream_tao_apps · GitHub.

Okay thanks @Morganh

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.