Pre-Processing the Original Dataset

[Dengj618]

hi @magehrig, There were some problems when I ran the preprocess_datasets.py file, and the data I generated was a little different from what you provided.
Such as

1. labels_v2\ timestamps_us.npy. There is a slight difference between the data shape and the value of this file. Maybe the difference doesn't matter.
   

2. The difference between timestamps_us.npy in the event representations v2 folder may be important.
   

Could you explain the meaning of the data under the event_representations_v2 and labels_v3 folders? This would help me train my data set. Thanks!

[magehrig]

Hi @Dengj618

For the Gen4 dataset you should be able to exactly replicate with the provided script. For the Gen1 dataset you might get slightly different data because I had to refactor the script and it seems that you are looking at the Gen1 dataset.

event_representations_v2:

• event_representations.h5: Contains the event representations. The data is in format of (Time Channel Height Width).
• objframe_idx_2_repr_idx.npy: A map of label index to representation index. E.g. I have label 42 and I want to know to which event representation this label belongs to I can just do objframe_idx_2_repr_idx[42] which gives me the index belonging to the event representation.
• timestamps_us.npy: Timestamps of the event representations

labels_v2:

• labels.npz: The labels
• timestamps_us.npy: The timestamps belonging to these labels.