TESTING

[Choneke]

Thanks for the nice work @ruiyan1995 .

I use the model to train it in my dataset and the results very nice. I modified it so that I can get both performance individual action and group action (using PCTDM model).

But I would like to ask for your suggestion, which part/script should I consider/modify so that I can use my trained model for the testing dataset (which have no label but have only bounding box for each individual in each frame).

I want to use my trained models, to test it (not using validation data which have both boxes and laabels).

I arrange the frame as needed. and each individual in a frame has x, y, w, h value.

Thank you.

[ruiyan1995]

@Choneke Thanks for your attention!

You can apply the pre-trained model (CNN-LSTM used in Action-stage) to extract features for the test data and directly evaluate it in Activity-stage. So, no labels will be required.

[Choneke]

Thank you for your reply, what I mean in my question is, already I have a train, my model, in my dataset and do an evaluation ( on a given validation data) and got good results both action and activity.

But now I have new data (of the same type), but It doesn't have an action label and activity label. It has only a frame number, x, y, w, h for each in a frame.

So what part should I consider so that to test these new data (neither in training nor validation) which have no action label and activity label, in order to get those labels using our action-trained model and activity trained model respectively.

Thank you

[ruiyan1995]

@Choneke Did you not know how to revise the code? In fact, the model doesn't require any label for test data (we only use the labels from validation data to report the results, but the model doesn't need them). You can revise the 'evaluation part' via outputting the predictions directly and do not compute the acc/mean_acc between predictions and labels that you don't have! If I misunderstand you, please let me know.

[Choneke]

@ruiyan1995 Thank you, I see it. I have to do some modifications to Solver.py file. And the way on how data preprocessed.

Thank you for your quick reply.