dtp statistics, targets, features

[MohamedHesham02]

Importing weights problem is solved
I want to know how can I apply the weights on my own rgb videos and if this is the replacement of these npy file ?

[olly-styles]

To use a custom dataset, you'll need to create your own versions of these files on the target dataset.

Statistics: Bounding box statistics, as described in Section 5.1 of the paper
Targets: Target bounding boxes
Features: compute optical flow and compute DTP features as explained in the Training on a custom dataset part of the readme.

[MohamedHesham02]

Ok, but the generated npy files from the DTP model is a prediction and targets files only
There are no features or statistics
I read section 5.1 but there's no guide how this file is generated
And the same problem for features file

[MohamedHesham02]

Is there a mail or WhatsApp for faster contact
This is my email address: mo7ameda7med@live.com

[olly-styles]

The features need to be computed by DTP. The feature vector comes from the final layer of DTP before prediction and is a vector of length 2048.

Each past bounding box has 8 values associated with it as described in Section 5.1. These 8 values are the statistics.