[Model] Yolov5/v5lite/v6/v7/v7end2end: CUDA preprocessing (#370)

* add yolo cuda preprocessing

* cmake build cuda src

* yolov5 support cuda preprocessing

* yolov5 cuda preprocessing configurable

* yolov5 update get mat data api

* yolov5 check cuda preprocess args

* refactor cuda function name

* yolo cuda preprocess padding value configurable

* yolov5 release cuda memory

* cuda preprocess pybind api update

* move use_cuda_preprocessing option to yolov5 model

* yolov5lite cuda preprocessing

* yolov6 cuda preprocessing

* yolov7 cuda preprocessing

* yolov7_e2e cuda preprocessing

* remove cuda preprocessing in runtime option

* refine log and cmake variable name

* fix model runtime ptr type

Co-authored-by: Jason <jiangjiajun@baidu.com>

python/fastdeploy/vision/detection/contrib/yolov5.py

@@ -37,6 +37,9 @@ class YOLOv5(FastDeployModel):
         return self._model.predict(input_image, conf_threshold,
                                    nms_iou_threshold)
 
+    def use_cuda_preprocessing(self, max_image_size=3840 * 2160):
+        return self._model.use_cuda_preprocessing(max_image_size)
+
     @staticmethod
     def preprocess(input_image,
                    size=[640, 640],

python/fastdeploy/vision/detection/contrib/yolov5lite.py

@@ -37,6 +37,9 @@ class YOLOv5Lite(FastDeployModel):
         return self._model.predict(input_image, conf_threshold,
                                    nms_iou_threshold)
 
+    def use_cuda_preprocessing(self, max_image_size=3840 * 2160):
+        return self._model.use_cuda_preprocessing(max_image_size)
+
     # 一些跟YOLOv5Lite模型有关的属性封装
     # 多数是预处理相关，可通过修改如model.size = [1280, 1280]改变预处理时resize的大小（前提是模型支持）
     @property

python/fastdeploy/vision/detection/contrib/yolov6.py

@@ -37,6 +37,9 @@ class YOLOv6(FastDeployModel):
         return self._model.predict(input_image, conf_threshold,
                                    nms_iou_threshold)
 
+    def use_cuda_preprocessing(self, max_image_size=3840 * 2160):
+        return self._model.use_cuda_preprocessing(max_image_size)
+
     # 一些跟YOLOv6模型有关的属性封装
     # 多数是预处理相关，可通过修改如model.size = [1280, 1280]改变预处理时resize的大小（前提是模型支持）
     @property

python/fastdeploy/vision/detection/contrib/yolov7.py

@@ -37,6 +37,9 @@ class YOLOv7(FastDeployModel):
         return self._model.predict(input_image, conf_threshold,
                                    nms_iou_threshold)
 
+    def use_cuda_preprocessing(self, max_image_size=3840 * 2160):
+        return self._model.use_cuda_preprocessing(max_image_size)
+
     # 一些跟YOLOv7模型有关的属性封装
     # 多数是预处理相关，可通过修改如model.size = [1280, 1280]改变预处理时resize的大小（前提是模型支持）
     @property

python/fastdeploy/vision/detection/contrib/yolov7end2end_trt.py

@@ -36,6 +36,9 @@ class YOLOv7End2EndTRT(FastDeployModel):
     def predict(self, input_image, conf_threshold=0.25):
         return self._model.predict(input_image, conf_threshold)
 
+    def use_cuda_preprocessing(self, max_image_size=3840 * 2160):
+        return self._model.use_cuda_preprocessing(max_image_size)
+
     # 一些跟模型有关的属性封装
     # 多数是预处理相关，可通过修改如model.size = [1280, 1280]改变预处理时resize的大小（前提是模型支持）
     @property