Traceback error when trying to train model on different actions (using my own videos) about realtime-action-recognition HOT 6 CLOSED

Hi Felixchenfy,

I was able to successfully train the model on different actions that I have recorded but have received a traceback error when I run 'run_detector'. It appears to only do this when attempting to classify the modified actions (i.e. it will run fine until I do the action at which point, I receive the error below).

For reference, I have trained the model on only two actions ('come here' and 'wave'). I have also attempted to run it when I have only trained it on one of those actions and received the same error. I am wondering if the program is designed to have exactly nine actions, in which case I would need to add more. As I understand it, the error is due to a mismatch in array size. Not sure how to fix, I would appreciate any guidance :)

Traceback (most recent call last):
File "C:\Users\steve\Documents\UNI\2019\S2\EGH400-2\code\working\Realtime-Action-Recognition-master\src\run_detector.py", line 316, in
dict_id2label = multipeople_classifier.classify(dict_id2skeleton)
File "C:\Users\steve\Documents\UNI\2019\S2\EGH400-2\code\working\Realtime-Action-Recognition-master\src\run_detector.py", line 258, in classify
id2label[id] = classifier.predict(skeleton) # predict label
File "C:\Users\steve\Documents\UNI\2019\S2\EGH400-2\code\working\Realtime-Action-Recognition-master\src\mylib\action_classifier.py", line 65, in predict
self.scores = self.smooth_scores(curr_scores)
File "C:\Users\steve\Documents\UNI\2019\S2\EGH400-2\code\working\Realtime-Action-Recognition-master\src\mylib\action_classifier.py", line 86, in smooth_scores
score_sums += score
ValueError: operands could not be broadcast together with shapes (9,) (2,) (9,)

from realtime-action-recognition.

@sbickley Hi, did you copy the action labels to the run_detector.py?

The detailed procedures are as follows:

When running src/train.ipynb, the function load_my_data should print out the action labels (computed from the names of training data):

And then, you need to copy the labels to the variable action_labels in run_detector.py, and then the program should work.

from realtime-action-recognition.

That is exactly what the problem was!!! Thanks @felixchenfy :)

from realtime-action-recognition.

@felixchenfy just wondering if there was any particular reason you chose to use jupyter notebook for the training stage?

from realtime-action-recognition.

@sbickley No particular reason. It's just that jupyter is easy to debug.

from realtime-action-recognition.

@sbickley Hi, I just refactored the code to make it more readable, as well as making the API easier to use.

The training code is now in the .py instead of jupyter notebook.

The classes are now set in a configuration file config/config.yaml.

Just for your information. Thanks!

from realtime-action-recognition.