Spaces:
Runtime error
Runtime error
File size: 12,040 Bytes
2366e36 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 |
# Copyright (c) OpenMMLab. All rights reserved.
import os.path as osp
import warnings
from typing import Any, Iterable
import numpy as np
import torch
from mmdet.models.builder import DETECTORS
from mmocr.models.textdet.detectors.single_stage_text_detector import \
SingleStageTextDetector
from mmocr.models.textdet.detectors.text_detector_mixin import \
TextDetectorMixin
from mmocr.models.textrecog.recognizer.encode_decode_recognizer import \
EncodeDecodeRecognizer
def inference_with_session(sess, io_binding, input_name, output_names,
input_tensor):
device_type = input_tensor.device.type
device_id = input_tensor.device.index
device_id = 0 if device_id is None else device_id
io_binding.bind_input(
name=input_name,
device_type=device_type,
device_id=device_id,
element_type=np.float32,
shape=input_tensor.shape,
buffer_ptr=input_tensor.data_ptr())
for name in output_names:
io_binding.bind_output(name)
sess.run_with_iobinding(io_binding)
pred = io_binding.copy_outputs_to_cpu()
return pred
@DETECTORS.register_module()
class ONNXRuntimeDetector(TextDetectorMixin, SingleStageTextDetector):
"""The class for evaluating onnx file of detection."""
def __init__(self,
onnx_file: str,
cfg: Any,
device_id: int,
show_score: bool = False):
if 'type' in cfg.model:
cfg.model.pop('type')
SingleStageTextDetector.__init__(self, **(cfg.model))
TextDetectorMixin.__init__(self, show_score)
import onnxruntime as ort
# get the custom op path
ort_custom_op_path = ''
try:
from mmcv.ops import get_onnxruntime_op_path
ort_custom_op_path = get_onnxruntime_op_path()
except (ImportError, ModuleNotFoundError):
warnings.warn('If input model has custom op from mmcv, \
you may have to build mmcv with ONNXRuntime from source.')
session_options = ort.SessionOptions()
# register custom op for onnxruntime
if osp.exists(ort_custom_op_path):
session_options.register_custom_ops_library(ort_custom_op_path)
sess = ort.InferenceSession(onnx_file, session_options)
providers = ['CPUExecutionProvider']
options = [{}]
is_cuda_available = ort.get_device() == 'GPU'
if is_cuda_available:
providers.insert(0, 'CUDAExecutionProvider')
options.insert(0, {'device_id': device_id})
sess.set_providers(providers, options)
self.sess = sess
self.device_id = device_id
self.io_binding = sess.io_binding()
self.output_names = [_.name for _ in sess.get_outputs()]
for name in self.output_names:
self.io_binding.bind_output(name)
self.cfg = cfg
def forward_train(self, img, img_metas, **kwargs):
raise NotImplementedError('This method is not implemented.')
def aug_test(self, imgs, img_metas, **kwargs):
raise NotImplementedError('This method is not implemented.')
def extract_feat(self, imgs):
raise NotImplementedError('This method is not implemented.')
def simple_test(self,
img: torch.Tensor,
img_metas: Iterable,
rescale: bool = False):
onnx_pred = inference_with_session(self.sess, self.io_binding, 'input',
self.output_names, img)
onnx_pred = torch.from_numpy(onnx_pred[0])
if len(img_metas) > 1:
boundaries = [
self.bbox_head.get_boundary(*(onnx_pred[i].unsqueeze(0)),
[img_metas[i]], rescale)
for i in range(len(img_metas))
]
else:
boundaries = [
self.bbox_head.get_boundary(*onnx_pred, img_metas, rescale)
]
return boundaries
@DETECTORS.register_module()
class ONNXRuntimeRecognizer(EncodeDecodeRecognizer):
"""The class for evaluating onnx file of recognition."""
def __init__(self,
onnx_file: str,
cfg: Any,
device_id: int,
show_score: bool = False):
if 'type' in cfg.model:
cfg.model.pop('type')
EncodeDecodeRecognizer.__init__(self, **(cfg.model))
import onnxruntime as ort
# get the custom op path
ort_custom_op_path = ''
try:
from mmcv.ops import get_onnxruntime_op_path
ort_custom_op_path = get_onnxruntime_op_path()
except (ImportError, ModuleNotFoundError):
warnings.warn('If input model has custom op from mmcv, \
you may have to build mmcv with ONNXRuntime from source.')
session_options = ort.SessionOptions()
# register custom op for onnxruntime
if osp.exists(ort_custom_op_path):
session_options.register_custom_ops_library(ort_custom_op_path)
sess = ort.InferenceSession(onnx_file, session_options)
providers = ['CPUExecutionProvider']
options = [{}]
is_cuda_available = ort.get_device() == 'GPU'
if is_cuda_available:
providers.insert(0, 'CUDAExecutionProvider')
options.insert(0, {'device_id': device_id})
sess.set_providers(providers, options)
self.sess = sess
self.device_id = device_id
self.io_binding = sess.io_binding()
self.output_names = [_.name for _ in sess.get_outputs()]
for name in self.output_names:
self.io_binding.bind_output(name)
self.cfg = cfg
def forward_train(self, img, img_metas, **kwargs):
raise NotImplementedError('This method is not implemented.')
def aug_test(self, imgs, img_metas, **kwargs):
if isinstance(imgs, list):
for idx, each_img in enumerate(imgs):
if each_img.dim() == 3:
imgs[idx] = each_img.unsqueeze(0)
imgs = imgs[0] # avoid aug_test
img_metas = img_metas[0]
else:
if len(img_metas) == 1 and isinstance(img_metas[0], list):
img_metas = img_metas[0]
return self.simple_test(imgs, img_metas=img_metas)
def extract_feat(self, imgs):
raise NotImplementedError('This method is not implemented.')
def simple_test(self,
img: torch.Tensor,
img_metas: Iterable,
rescale: bool = False):
"""Test function.
Args:
imgs (torch.Tensor): Image input tensor.
img_metas (list[dict]): List of image information.
Returns:
list[str]: Text label result of each image.
"""
onnx_pred = inference_with_session(self.sess, self.io_binding, 'input',
self.output_names, img)
onnx_pred = torch.from_numpy(onnx_pred[0])
label_indexes, label_scores = self.label_convertor.tensor2idx(
onnx_pred, img_metas)
label_strings = self.label_convertor.idx2str(label_indexes)
# flatten batch results
results = []
for string, score in zip(label_strings, label_scores):
results.append(dict(text=string, score=score))
return results
@DETECTORS.register_module()
class TensorRTDetector(TextDetectorMixin, SingleStageTextDetector):
"""The class for evaluating TensorRT file of detection."""
def __init__(self,
trt_file: str,
cfg: Any,
device_id: int,
show_score: bool = False):
if 'type' in cfg.model:
cfg.model.pop('type')
SingleStageTextDetector.__init__(self, **(cfg.model))
TextDetectorMixin.__init__(self, show_score)
from mmcv.tensorrt import TRTWrapper, load_tensorrt_plugin
try:
load_tensorrt_plugin()
except (ImportError, ModuleNotFoundError):
warnings.warn('If input model has custom op from mmcv, \
you may have to build mmcv with TensorRT from source.')
model = TRTWrapper(
trt_file, input_names=['input'], output_names=['output'])
self.model = model
self.device_id = device_id
self.cfg = cfg
def forward_train(self, img, img_metas, **kwargs):
raise NotImplementedError('This method is not implemented.')
def aug_test(self, imgs, img_metas, **kwargs):
raise NotImplementedError('This method is not implemented.')
def extract_feat(self, imgs):
raise NotImplementedError('This method is not implemented.')
def simple_test(self,
img: torch.Tensor,
img_metas: Iterable,
rescale: bool = False):
with torch.cuda.device(self.device_id), torch.no_grad():
trt_pred = self.model({'input': img})['output']
if len(img_metas) > 1:
boundaries = [
self.bbox_head.get_boundary(*(trt_pred[i].unsqueeze(0)),
[img_metas[i]], rescale)
for i in range(len(img_metas))
]
else:
boundaries = [
self.bbox_head.get_boundary(*trt_pred, img_metas, rescale)
]
return boundaries
@DETECTORS.register_module()
class TensorRTRecognizer(EncodeDecodeRecognizer):
"""The class for evaluating TensorRT file of recognition."""
def __init__(self,
trt_file: str,
cfg: Any,
device_id: int,
show_score: bool = False):
if 'type' in cfg.model:
cfg.model.pop('type')
EncodeDecodeRecognizer.__init__(self, **(cfg.model))
from mmcv.tensorrt import TRTWrapper, load_tensorrt_plugin
try:
load_tensorrt_plugin()
except (ImportError, ModuleNotFoundError):
warnings.warn('If input model has custom op from mmcv, \
you may have to build mmcv with TensorRT from source.')
model = TRTWrapper(
trt_file, input_names=['input'], output_names=['output'])
self.model = model
self.device_id = device_id
self.cfg = cfg
def forward_train(self, img, img_metas, **kwargs):
raise NotImplementedError('This method is not implemented.')
def aug_test(self, imgs, img_metas, **kwargs):
if isinstance(imgs, list):
for idx, each_img in enumerate(imgs):
if each_img.dim() == 3:
imgs[idx] = each_img.unsqueeze(0)
imgs = imgs[0] # avoid aug_test
img_metas = img_metas[0]
else:
if len(img_metas) == 1 and isinstance(img_metas[0], list):
img_metas = img_metas[0]
return self.simple_test(imgs, img_metas=img_metas)
def extract_feat(self, imgs):
raise NotImplementedError('This method is not implemented.')
def simple_test(self,
img: torch.Tensor,
img_metas: Iterable,
rescale: bool = False):
"""Test function.
Args:
imgs (torch.Tensor): Image input tensor.
img_metas (list[dict]): List of image information.
Returns:
list[str]: Text label result of each image.
"""
with torch.cuda.device(self.device_id), torch.no_grad():
trt_pred = self.model({'input': img})['output']
label_indexes, label_scores = self.label_convertor.tensor2idx(
trt_pred, img_metas)
label_strings = self.label_convertor.idx2str(label_indexes)
# flatten batch results
results = []
for string, score in zip(label_strings, label_scores):
results.append(dict(text=string, score=score))
return results
|