[CANN] Add Ascend NPU support

examples/person_detection/scrfd_person.py

@@ -11,8 +11,8 @@
 
 def detect_person(img, detector):
  bboxes, kpss = detector.detect(img)
- bboxes = np.round(bboxes[:,:4]).astype(np.int)
- kpss = np.round(kpss).astype(np.int)
+ bboxes = np.round(bboxes[:,:4]).astype(np.int32)
+ kpss = np.round(kpss).astype(np.int32)
  kpss[:,:,0] = np.clip(kpss[:,:,0], 0, img.shape[1])
  kpss[:,:,1] = np.clip(kpss[:,:,1], 0, img.shape[0])
  vbboxes = bboxes.copy()

python-package/README.md

@@ -14,7 +14,7 @@ For ``insightface<=0.1.5``, we use MXNet as inference backend.
 
 Starting from insightface>=0.2, we use onnxruntime as inference backend.
 
-You have to install ``onnxruntime-gpu`` manually to enable GPU inference, or install ``onnxruntime`` to use CPU only inference.
+You have to install ``onnxruntime-gpu`` manually to enable GPU inference, or ``pip install onnxruntime-cann`` manually to enable NPU inference, or install ``onnxruntime`` to use CPU only inference.
 
 ## Change Log
 
@@ -54,7 +54,7 @@ import insightface
 from insightface.app import FaceAnalysis
 from insightface.data import get_image as ins_get_image
 
-app = FaceAnalysis(providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])
+app = FaceAnalysis(providers=['CUDAExecutionProvider', 'CANNExecutionProvider', 'CPUExecutionProvider'])
 app.prepare(ctx_id=0, det_size=(640, 640))
 img = ins_get_image('t1')
 faces = app.get(img)

python-package/insightface/model_zoo/model_zoo.py

@@ -68,7 +68,7 @@ def find_onnx_file(dir_path):
  return paths[-1]
 
 def get_default_providers():
- return ['CUDAExecutionProvider', 'CPUExecutionProvider']
+ return ['CUDAExecutionProvider', 'CANNExecutionProvider', 'CPUExecutionProvider']
 
 def get_default_provider_options():
  return None

web-demos/src_recognition/arcface_onnx.py

@@ -42,7 +42,7 @@ def __init__(self, model_file=None, session=None):
  self.input_std = input_std
  #print('input mean and std:', self.input_mean, self.input_std)
  if self.session is None:
- self.session = onnxruntime.InferenceSession(self.model_file, providers=['CUDAExecutionProvider'])
+ self.session = onnxruntime.InferenceSession(self.model_file, providers=['CUDAExecutionProvider', 'CANNExecutionProvider'])
  input_cfg = self.session.get_inputs()[0]
  input_shape = input_cfg.shape
  input_name = input_cfg.name

web-demos/src_recognition/scrfd.py

@@ -74,7 +74,7 @@ def __init__(self, model_file=None, session=None):
  if self.session is None:
  assert self.model_file is not None
  assert osp.exists(self.model_file)
- self.session = onnxruntime.InferenceSession(self.model_file, providers=['CUDAExecutionProvider'])
+ self.session = onnxruntime.InferenceSession(self.model_file, providers=['CUDAExecutionProvider', 'CANNExecutionProvider'])
  self.center_cache = {}
  self.nms_thresh = 0.4
  self.det_thresh = 0.5