HI there iam currently working on Fine-tuning of PoseClassificationNet and currently i have seperated videos of different poses like walking,standing etc now i have also run the dataset convert command for it but the thing is now i have multiple numpy arrays. But in training notebook the dataset is passed as train.numpy, test.numpy etc, like only these two files but i have multiple videos converted to numpy arrays,collectively 2 walking and 1 standing videos json have been passed and how to deal with this?
2- I also have to generate pickle files similarly for these videos so let me know how to do that too
What do you use for the finetuning? TAO-toolkit?
Yes it is Tao Toolkit
To combine multiple NumPy arrays into a single array for training and testing, you can use the np.concatenate() function. This will allow you to merge all your walking and standing videos into two main arrays: one for training and one for testing.
Decide on a ratio for splitting your data into training and testing sets. A common approach is to use 80% for training and 20% for testing. You can use np.random.shuffle() to randomize the order of your data before splitting.
Ensure that each NumPy array has a corresponding action label (e.g., walking or standing). You can store these labels in separate arrays or as part of a dictionary with the NumPy arrays.
For Pickle Format, the pickle files should contain both the keypoints data (as NumPy arrays) and their corresponding action labels.
Pleas use Python’s pickle module to save your merged NumPy arrays and labels into pickle files.
An example:
20250308_forum_326005.txt (1.2 KB)