box_processing.py load_box_event function simplification

[grygry12345]

I checked your box_processing_differance.py. I find out the code complicated. So, I simplified.

def load_box_events(metadata, batch_start_time, duration):
    ending = metadata.get_ending()
    box_path = '_bbox.npy'.join(metadata.path.rsplit(ending, 1))
    box_events = np.load(box_path)

    # Filter out the events that are outside the desired time frame
    box_events = box_events[(box_events['t'] >= batch_start_time) & (box_events['t'] < batch_start_time + duration)]

    # Initialize the list to hold output event data
    out_list = []

    # Define the new class_ids for the features
    class_ids = {
        'face_bbox': 0, # Assuming 'face_bbox' is for face
        'x0y0': 1,      # Assuming 'x0y0' is for lip
        'x1y1': 2,      # Assuming 'x1y1' is for eye1
        'x2y2': 3,      # Assuming 'x2y2' is for eye2
        'x3y3': 4,      # Assuming 'x3y3' is for nose
        'x4y4': 5,      # Assuming 'x4y4' is for lips
    }

    # Define the dtype for the structured array
    out_dtype = np.dtype([
        ('t', '<i8'), ('x', '<f4'), ('y', '<f4'), ('w', '<f4'), ('h', '<f4'),
        ('class_id', '<u4'), ('track_id', '<u4'), ('class_confidence', '<f4')
    ])

    # Process each event and create a separate record for each feature
    for event in box_events:
        # First add the face bbox event
        face_bbox_event = (event['t'], event['x'], event['y'], event['w'], event['h'], class_ids['face_bbox'], event['track_id'], event['class_confidence'])
        out_list.append(face_bbox_event)
        
        # Now add the feature events
        for feature in ['x0y0', 'x1y1', 'x2y2', 'x3y3', 'x4y4']:
            class_id = class_ids[feature]
            x_key, y_key = feature[:2], feature[2:]
            feature_event = (event['t'], 
                             event[x_key] - ((event['w'] / 3) / 2), 
                             event[y_key] - ((event['h'] / 4) / 2),
                             event['w'] / 3, 
                             event['h'] / 4, 
                             class_id, 
                             event['track_id'], 
                             event['class_confidence'])
            out_list.append(feature_event)

    # Now convert out_list to a structured array
    out = np.array(out_list, dtype=out_dtype)

    return out