heigth, width = img.shape[:2] AttributeError: 'NoneType' object has no attribute 'shape'

在Python中，当你尝试执行 `height, width = img.shape[:2]` 这行代码时，会引发一个 `AttributeError` 错误，提示 `'NoneType' object has no attribute 'shape'`。这个错误表明你正在尝试访问 `img` 对象的 `shape` 属性，但是 `img` 是 `None` 或者没有这个属性。 通常情况下，`img` 应该是一个图像对象，比如在使用图像处理库（如OpenCV、PIL等）时，`img` 可能是一个加载的图像对象。这些库中加载图像的函数通常会返回一个图像对象，这个对象包含了图像的元数据，比如尺寸（通过 `shape` 属性）。 可能的情况是 `img` 没有成功加载到任何图像，或者它根本不是一个图像对象，而是一个 `None` 类型。这可能是由于以下原因造成的： 1. 图像文件路径错误或文件损坏，导致无法加载图像。 2. 图像处理库的函数调用不正确，例如参数错误，导致返回了 `None`。 3. 在加载图像之前，变量 `img` 被错误地赋值为 `None`。 为了解决这个问题，你需要检查图像加载的代码部分，确保图像文件路径正确，且图像处理库被正确使用。此外，可以添加一些错误检查的代码，以确保在尝试访问 `shape` 属性之前，`img` 不是 `None`。 示例代码： ```python if img is not None: height, width = img.shape[:2] else: print("图像加载失败，请检查图像路径或文件状态。") ```

从错误信息来看，问题出在 `img` 是 `None`。这通常意味着在调用 `_preprocess_frame_gpu` 或 `_letterbox_gpu` 方法时，传入的图像数据为 `None`。我们需要确保在调用这些方法之前，图像已经正确加载。 以下是改进后的代码，添加了对图像加载失败的检查，并确保不会传递 `None` 给预处理函数： ```python import cv2 import numpy as np import pycuda.driver as cuda import pycuda.gpuarray as gpuarray from pycuda.compiler import SourceModule class TRTEngine: def __init__(self, engine_path): # 初始化部分保持不变 self.ctx_lock = threading.Lock() self.cfx = cuda.Device(0).make_context() self.stream = cuda.Stream() self.logger = trt.Logger(trt.Logger.WARNING) self._register_plugins() self.engine = self._load_engine(engine_path) if self.engine is None: raise RuntimeError("Failed to load TensorRT engine") self.context = self.engine.create_execution_context() # 输入参数初始化 self.input_name = None self.input_shape = None for binding in self.engine: if self.engine.get_tensor_mode(binding) == trt.TensorIOMode.INPUT: self.input_name = binding if self.engine.has_implicit_batch_dimension: self.input_shape = self.engine.get_tensor_shape(binding)[1:] else: self.input_shape = self.engine.get_tensor_profile_shape(binding, 0)[1][1:] break if self.input_name is None or self.input_shape is None: raise RuntimeError("Could not find input binding") print(f"模型输入名称: {self.input_name}, 输入形状: {self.input_shape}") if self.engine.has_implicit_batch_dimension: self.context.set_binding_shape(0, (1,) + self.input_shape) else: self.context.set_input_shape(self.input_name, (1,) + self.input_shape) self._setup_bindings() def _letterbox_gpu(self, img, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True, stride=32): """GPU加速的letterbox函数""" if img is None: raise ValueError("Input image is None. Please ensure the image is loaded correctly.") shape = img.shape[:2] if isinstance(new_shape, int): new_shape = (new_shape, new_shape) new_h, new_w = new_shape r = min(new_h / shape[0], new_w / shape[1]) if not scaleup: r = min(r, 1.0) scaled_w = int(round(shape[1] * r)) scaled_h = int(round(shape[0] * r)) dw = new_w - scaled_w dh = new_h - scaled_h top, bottom = dh // 2, dh - (dh // 2) left, right = dw // 2, dw - (dw // 2) # 将图像上传到GPU img_gpu = gpuarray.to_gpu(img.astype(np.float32)) # 缩放图像 if (scaled_w, scaled_h) != (shape[1], shape[0]): img_resized = cv2.cuda.resize(img_gpu, (scaled_w, scaled_h), interpolation=cv2.INTER_LINEAR) else: img_resized = img_gpu # 创建填充后的图像 result = np.ones((new_h, new_w, 3), dtype=np.uint8) * color[0] result[top:top+scaled_h, left:left+scaled_w] = img_resized.get() return result, (r, r), (dw, dh) def infer(self, frame): with self.ctx_lock: self.cfx.push() try: if frame is None: raise ValueError("Input frame is None. Please ensure the frame is loaded correctly.") input_data = self._preprocess_frame_gpu(frame) expected_size = trt.volume((1, *self.input_shape)) if input_data.size != expected_size: raise ValueError(f"输入数据大小不匹配：{input_data.size} vs {expected_size}") np.copyto(self.inputs[0]['host'], input_data.ravel()) for inp in self.inputs: cuda.memcpy_htod_async(inp['device'], inp['host'], self.stream) if self.engine.has_implicit_batch_dimension: self.context.execute_async_v2( bindings=self.bindings, stream_handle=self.stream.handle ) else: self.context.set_input_shape(self.input_name, (1,) + self.input_shape) for out in self.outputs: out['shape'] = self.context.get_tensor_shape(out['name']) out['host'] = cuda.pagelocked_empty(trt.volume(out['shape']), out['dtype']) for out in self.outputs: self.context.set_tensor_address(out['name'], int(out['device'])) for inp in self.inputs: self.context.set_tensor_address(inp['name'], int(inp['device'])) self.context.execute_async_v3(stream_handle=self.stream.handle) outputs = [] for out in self.outputs: output = np.empty(out['shape'], dtype=out['dtype']) cuda.memcpy_dtoh_async(output, out['device'], self.stream) outputs.append(output) self.stream.synchronize() return self._postprocess(outputs) finally: self.cfx.pop() def _preprocess_frame_gpu(self, img, size=(640, 640)): """GPU加速的视频帧预处理方法""" if img is None: raise ValueError("Input image is None. Please ensure the image is loaded correctly.") orig_height, orig_width = img.shape[:2] # 执行GPU加速的letterbox处理 img, self.ratio, (self.dw, self.dh) = self._letterbox_gpu(img, new_shape=size, stride=32) # 将图像上传到GPU img_gpu = cuda.np_to_array(img, order='C') # 颜色空间转换：BGR -> RGB mod = SourceModule(""" __global__ void bgr_to_rgb(float *img, int height, int width) { int idx = threadIdx.x + blockIdx.x * blockDim.x; if (idx < height * width) { int b_idx = idx * 3; int r_idx = b_idx + 2; float temp = img[b_idx]; img[b_idx] = img[r_idx]; img[r_idx] = temp; } } """) bgr_to_rgb = mod.get_function("bgr_to_rgb") bgr_to_rgb(img_gpu, np.int32(orig_height), np.int32(orig_width), block=(1024, 1, 1), grid=(int(np.ceil(orig_height * orig_width / 1024)), 1)) # 归一化：将像素值从[0, 255]映射到[0, 1] img_gpu /= 255.0 # 调整维度顺序：HWC -> CHW img_gpu = img_gpu.transpose((2, 0, 1)).astype(np.float32) return img_gpu.get() def _postprocess(self, outputs): if len(outputs) < 4: return [] num_detections = int(outputs[0][0][0]) boxes = outputs[1][0][:num_detections] scores = outputs[2][0][:num_detections] class_ids = outputs[3][0][:num_detections].astype(int) boxes[:, [0, 2]] = (boxes[:, [0, 2]] - self.dw // 2) / self.ratio[0] boxes[:, [1, 3]] = (boxes[:, [1, 3]] - self.dh // 2) / self.ratio[1] boxes[:, [0, 2]] = np.clip(boxes[:, [0, 2]], 0, self.orig_width) boxes[:, [1, 3]] = np.clip(boxes[:, [1, 3]], 0, self.orig_height) return [{ "box": box.astype(int).tolist(), "score": float(score), "class_id": int(cls_id) } for box, score, cls_id in zip(boxes, scores, class_ids)] ``` ### 解释 1. **检查图像是否为 `None`**： - 在 `_letterbox_gpu` 和 `_preprocess_frame_gpu` 方法中，首先检查 `img` 是否为 `None`。 - 如果是 `None`，抛出一个 `ValueError` 异常，提示用户图像未正确加载。 2. **确保图像正确加载**： - 在 `infer` 方法中，同样检查 `frame` 是否为 `None`，以确保在推理之前图像已正确加载。 ###