o ॵi@sddlmZmZmZmZddlmZddlmZddl m Z ddl m Z ddl mZddlmZddlmZeZe jejejd Gd d d e Zd S) )AnyDictListUnion) Pipelines)Model)Pipeline) PIPELINES)Tasks)Image) get_logger) module_namecseZdZdZdeffdd Zdeeeefffdd Z dede ee ffd d Z de ee fde ee ffd d Z d e ee fde ee ffddZZS)CardDetectionPipelinea  Card Detection Pipeline. Examples: >>> from modelscope.pipelines import pipeline >>> detector = pipeline('card-detection', 'damo/cv_resnet_carddetection_scrfd34gkps') >>> detector("http://www.modelscope.cn/api/v1/models/damo/cv_resnet_carddetection_scrfd34gkps/repo?Revision=master" >>> "&FilePath=description/card_detection1.jpg") >>> { >>> "boxes": [ >>> [ >>> 446.9007568359375, >>> 36.374977111816406, >>> 907.0919189453125, >>> 337.439208984375 >>> ], >>> [ >>> 454.3310241699219, >>> 336.08477783203125, >>> 921.26904296875, >>> 641.7871704101562 >>> ] >>> ], >>> "keypoints": [ >>> [ >>> 457.34710693359375, >>> 339.02044677734375, >>> 446.72271728515625, >>> 52.899078369140625, >>> 902.8200073242188, >>> 35.063236236572266, >>> 908.5877685546875, >>> 325.62030029296875 >>> ], >>> [ >>> 465.2864074707031, >>> 642.8411254882812, >>> 454.38568115234375, >>> 357.4076232910156, >>> 902.5343017578125, >>> 334.18377685546875, >>> 922.0982055664062, >>> 621.0704345703125 >>> ] >>> ], >>> "scores": [ >>> 0.9296008944511414, >>> 0.9260380268096924 >>> ] >>> } >>> modelc sBtjdd|i|t|jtsJd|j|j}||_dS)z use `model` to create a face detection pipeline for prediction Args: model: model id on modelscope hub or `ScrfdDetect` Model. preprocessor: `SCRFDPreprocessor`. rz model object is not initialized.N)super__init__ isinstancerrtodevicedetector)selfrkwargsr __class__rc/home/ubuntu/.local/lib/python3.10/site-packages/modelscope/pipelines/cv/card_detection_pipeline.pyrHs zCardDetectionPipeline.__init__inputc stj|fi|S)a Detect objects (bounding boxes or keypoints) in the image(s) passed as inputs. Args: input (`Image` or `List[Image]`): The pipeline handles three types of images: - A string containing an HTTP(S) link pointing to an image - A string containing a local path to an image - An image loaded in PIL or opencv directly The pipeline accepts either a single image or a batch of images. Images in a batch must all be in the same format. Return: A dictionary of result or a list of dictionary of result. If the input is an image, a dictionary is returned. If input is a list of image, a list of dictionary is returned. The dictionary contain the following keys: - **scores** (`List[float]`) -- The detection score for each card in the image. - **boxes** (`List[float]) -- The bounding boxe [x1, y1, x2, y2] of detected objects in in image's original size. - **keypoints** (`List[Dict[str, int]]`, optional) -- The corner kepoint [x1, y1, x2, y2, x3, y3, x4, y4] of detected object in image's original size. )r__call__)rrrrrrrUszCardDetectionPipeline.__call__returncCs`||}d|vr.ddlm}m}||gdd}t|jjr.||t|jjgd}|S)N img_metasr)collatescatter)samples_per_gpu) preprocessor mmcv.parallelr r!nextr parametersis_cudar)rrresultr r!rrr preprocessss z CardDetectionPipeline.preprocesscCs|jdi|S)Nr)r)rrrrrforwardszCardDetectionPipeline.forwardinputscCs|S)Nr)rr,rrr postprocesssz!CardDetectionPipeline.postprocess)__name__ __module__ __qualname____doc__strrrr rrrrr*r+r- __classcell__rrrrrs6 " *rN)typingrrrrmodelscope.metainfor!modelscope.models.base.base_modelrmodelscope.pipelines.basermodelscope.pipelines.builderr modelscope.utils.constantr $modelscope.utils.input_output_typingr modelscope.utils.loggerr loggerregister_modulecard_detectionrrrrrs