5.24. sail.algo_yolov5_post_3output

For post-processing interfaces with a triple output YOLOv5 model, internally implemented using thread pools.

5.24.1. __init__

Interface:

def __init__(
            self,
            shape: list[list[int]],
            network_w:int = 640,
            network_h:int = 640,
            max_queue_size: int=20)

Parameters:

• shape: list[list[int]]

The shape of the input data.

• network_w: int

The input width of the model, which defaults to 640.

• network_h: int

The input height of the model, which defaults to 640.

• max_queue_size: int

Maximum length of cached data.

5.24.2. push_data

Support input with arbitrary batchsize.

Interface:

def push_data(self,
    channel_idx: list[int],
    image_idx: list[int],
    input_data: list[TensorPTRWithName],
    dete_threshold: list[float],
    nms_threshold: list[float],
    ost_w: list[int],
    ost_h: list[int],
    padding_attrs: list[list[int]]) -> int

Parameters:

• channel_idx: list[int]

The channel number of the input image.

• image_idx: list[int]

The sequence number of the input image.

• input_data: list[TensorPTRWithName],

The input data, including three outputs.

• dete_threshold: list[float]

Detection threshold sequence.

• nms_threshold: list[float]

nms threshold.

• ost_w: list[int]

The width of original image.

• ost_h: list[int]

The height of original image.

• padding_attrs: list[list[int]]

The attribute list of the fill image, starting point coordinate x, starting point coordinate y, width after scaling, height after scaling.

Returns:

Return 0 if successful, otherwise failed.

5.24.3. get_result_npy

Get the final detection result.

Interface:

def get_result_npy(self)
        -> tuple[tuple[int, int, int, int, int, float],int, int]

Returns: tuple[tuple[left, top, right, bottom, class_id, score],channel_idx, image_idx]

• left: int

The left x coordinate of the detection result.

• top: int

The top y coordinate of the detection result.

• right: int

The right x coordinate of the detection result.

• bottom: int

The bottom y coordinate of the detection result.

• class_id: int

Category number of detection result.

• score: float

Score of detection result.

• channel_idx: int

The channel index of original image.

• image_idx: int

The image index of original image.

5.24.4. reset_anchors

Reset anchors.

Interface:

def reset_anchors(self, anchors_new: list[list[list[int]]]) -> int

Parameters:

• anchors_new: list[list[list[int]]]

new anchors.

Returns:

Return 0 if successful, otherwise failed.