Image Classification(图像分类)#

fastdeploy.vision.classification.PaddleClasPreprocessor#

class fastdeploy.vision.classification.PaddleClasPreprocessor(config_file)[source]#

Create a preprocessor for PaddleClasModel from configuration file

Parameters

config_file – (str)Path of configuration file, e.g resnet50/inference_cls.yaml

disable_normalize()[source]#

This function will disable normalize in preprocessing step.

disable_permute()[source]#

This function will disable hwc2chw in preprocessing step.

initial_resize_on_cpu(v)[source]#

When the initial operator is Resize, and input image size is large, maybe it’s better to run resize on CPU, because the HostToDevice memcpy is time consuming. Set this True to run the initial resize on CPU. :param: v: True or False

run(input_ims)#

Process input image

Param

input_ims: (list of numpy.ndarray) The input images

Returns

list of FDTensor

use_cuda(enable_cv_cuda=False, gpu_id=-1)#

Use CUDA processors

Param

enable_cv_cuda: Ture: use CV-CUDA, False: use CUDA only

Param

gpu_id: GPU device id

fastdeploy.vision.classification.PaddleClasPostprocessor#

class fastdeploy.vision.classification.PaddleClasPostprocessor(topk=1)[source]#

Create a postprocessor for PaddleClasModel

Parameters

topk – (int)Filter the top k classify label

run(runtime_results)[source]#

Postprocess the runtime results for PaddleClasModel

Param

runtime_results: (list of FDTensor)The output FDTensor results from runtime

Returns

list of ClassifyResult(If the runtime_results is predict by batched samples, the length of this list equals to the batch size)

fastdeploy.vision.classification.PaddleClasModel#

class fastdeploy.vision.classification.PaddleClasModel(model_file, params_file, config_file, runtime_option=None, model_format=<ModelFormat.PADDLE: 1>)[source]#

Load a image classification model exported by PaddleClas.

Parameters

  • model_file – (str)Path of model file, e.g resnet50/inference.pdmodel

  • params_file – (str)Path of parameters file, e.g resnet50/inference.pdiparams, if the model_fomat is ModelFormat.ONNX, this param will be ignored, can be set as empty string

  • config_file – (str) Path of configuration file for deploy, e.g resnet50/inference_cls.yaml

  • runtime_option – (fastdeploy.RuntimeOption)RuntimeOption for inference this model, if it’s None, will use the default backend on CPU

  • model_format – (fastdeploy.ModelForamt)Model format of the loaded model

batch_predict(images)[source]#

Classify a batch of input image

Parameters

im – (list of numpy.ndarray) The input image list, each element is a 3-D array with layout HWC, BGR format

:return list of ClassifyResult

clone()[source]#

Clone PaddleClasModel object

Returns

a new PaddleClasModel object

get_profile_time()#

Get profile time of Runtime after the profile process is done.

property postprocessor#

Get PaddleClasPostprocessor object of the loaded model

:return PaddleClasPostprocessor

predict(im, topk=1)[source]#

Classify an input image

Parameters

  • im – (numpy.ndarray) The input image data, a 3-D array with layout HWC, BGR format

  • topk – (int) Filter the topk classify result, default 1

Returns

ClassifyResult

property preprocessor#

Get PaddleClasPreprocessor object of the loaded model

:return PaddleClasPreprocessor

fastdeploy.vision.classification.YOLOv5Cls#

class fastdeploy.vision.classification.YOLOv5Cls(model_file, params_file='', runtime_option=None, model_format=<ModelFormat.ONNX: 2>)[source]#

Load a YOLOv5Cls model exported by YOLOv5Cls.

Parameters

  • model_file – (str)Path of model file, e.g ./YOLOv5Cls.onnx

  • params_file – (str)Path of parameters file, e.g yolox/model.pdiparams, if the model_fomat is ModelFormat.ONNX, this param will be ignored, can be set as empty string

  • runtime_option – (fastdeploy.RuntimeOption)RuntimeOption for inference this model, if it’s None, will use the default backend on CPU

  • model_format – (fastdeploy.ModelForamt)Model format of the loaded model

batch_predict(images)[source]#

Classify a batch of input image

Parameters

im – (list of numpy.ndarray) The input image list, each element is a 3-D array with layout HWC, BGR format

:return list of ClassifyResult

get_profile_time()#

Get profile time of Runtime after the profile process is done.

property postprocessor#

Get YOLOv5ClsPostprocessor object of the loaded model

:return YOLOv5ClsPostprocessor

predict(input_image)[source]#

Classify an input image

Parameters

input_image – (numpy.ndarray)The input image data, 3-D array with layout HWC, BGR format

Returns

ClassifyResult

property preprocessor#

Get YOLOv5ClsPreprocessor object of the loaded model

:return YOLOv5ClsPreprocessor

fastdeploy.vision.classification.ResNet#

class fastdeploy.vision.classification.ResNet(model_file, params_file='', runtime_option=None, model_format=<ModelFormat.ONNX: 2>)[source]#

Load a image classification model exported by torchvision.ResNet.

Parameters

  • model_file – (str)Path of model file, e.g resnet/resnet50.onnx

  • params_file – (str)Path of parameters file, if the model_fomat is ModelFormat.ONNX, this param will be ignored, can be set as empty string

  • runtime_option – (fastdeploy.RuntimeOption)RuntimeOption for inference this model, if it’s None, will use the default backend on CPU

  • model_format – (fastdeploy.ModelForamt)Model format of the loaded model, default is ONNX

get_profile_time()#

Get profile time of Runtime after the profile process is done.

property mean_vals#

Returns the mean value of normlization, default mean_vals = [0.485f, 0.456f, 0.406f];

predict(input_image, topk=1)[source]#

Classify an input image

Parameters

  • input_image – (numpy.ndarray)The input image data, 3-D array with layout HWC, BGR format

  • topk – (int)The topk result by the classify confidence score, default 1

Returns

ClassifyResult

property size#

Returns the preprocess image size, default size = [224, 224];

property std_vals#

Returns the std value of normlization, default std_vals = [0.229f, 0.224f, 0.225f];