= R"pydeepstream(
Extracts metadata from the batch into a list of dictionaries.
This method efficiently extracts all metadata (bounding boxes, labels, confidences, segmentation maps, audio labels, etc.)
from all frames in the batch. The extraction, data copying, and type conversions are performed in C++ for optimal performance.
:arg max_items: *int* - Maximum number of items in the returned list. The list will have this capacity,
indexed by frame *pad_index*.
:returns: *list of dict* - A list with size equal to *max_items*. Each element corresponds to a frame
indexed by *pad_index*. Elements without data will be *None*.
Each dictionary contains:
- **shape**: *list of int* - Frame dimensions [height, width]
- **bboxes**: *list of list of int* - Bounding boxes in [left, top, right, bottom] format
- **probs**: *list of float* - Confidence scores for each detected object
- **labels**: *list of list of str* - Labels for each object (including classifier outputs)
- **objects**: *list of int* - Object IDs for tracking
- **seg_maps**: *list of numpy.ndarray* - Segmentation maps (instance masks or semantic segmentation) as int arrays
- **timestamp**: *int* - Frame presentation timestamp (PTS)
Audio dictionaries also include **media_type**, **sample_rate**, **num_channels**,
**num_samples_per_frame**, **format**, **layout**, **infer_done**, **class_id**,
**class_label**, and **confidence**.
)pydeepstream"