Training New Actions with Pre-trained Models: Do Previous Classifications Persist?

Action Recognition Net | NVIDIA NGC
When I use a pre-trained model to train new actions, will the classifications of actions that the model had before training disappear?
If I want to retain the default five actions of this action recognition model, what should I do? I’ve already configured it as follows, but it’s not working. (The first five actions are from the pre-trained model, while smoking and drinking are ones I trained.)

dataset:
train_dataset_dir: /data/train
val_dataset_dir: /data/test
label_map:
push: 0
fall_floor: 1
walk: 2
run: 3
ride_bike: 4
smoke: 5
drink: 6

The dataset should contain these first five actions.

You can refer to notebook tao_tutorials/notebooks/tao_launcher_starter_kit/action_recognition_net/actionrecognitionnet.ipynb at main · NVIDIA/tao_tutorials · GitHub.

I just followed this tutorial. What I want to ask is how to train new data while retaining the pre-trained model. What should I modify?

If you want to retain the default five actions of this action recognition model, it is expected that the training dataset still contain the five classes.

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

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