使用训练后ResNet模型推理遇TypeError，求代码修正方案

错误场景 #

执行代码output = model([image_tensor])时触发TypeError，报错显示conv2d函数接收到无效参数组合，传入了列表而非Tensor类型数据。

完整代码 #

# import libraries
import torch
import cv2
import numpy as np

# load the trained ResNet model
from torch.autograd import Variable

model = torch.load("model.pth")
model.eval()

# load the image of the steel pipe
image = cv2.imread("img/75.jpg")
DEVICE = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

# preprocess the image
image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) # convert to RGB
image_resized = cv2.resize(image_rgb, (224, 224)) # resize to 224x224
image_normalized = image_resized / 255.0 # normalize to [0, 1]
image_tensor = torch.from_numpy(image_normalized) # convert to tensor
image_tensor = image_tensor.permute(2, 0, 1) # change the order of channels
image_tensor = image_tensor.unsqueeze(0) # add a batch dimension
data = Variable(image_tensor).to(DEVICE)

# pass the image tensor to the model and get the output
print(image)
output = model([image_tensor])

# get the predicted bounding boxes, labels, scores, and masks
boxes = output["boxes"].detach().numpy()
labels = output["labels"].detach().numpy()
scores = output["scores"].detach().numpy()
masks = output["masks"].detach().numpy()


# filter out the objects that are not steel pipes
threshold = 0.5 # you can adjust this value
indices = np.where((labels == 1) & (scores >= threshold))[0] # assuming label 1 is for steel pipe
boxes = boxes[indices]
masks = masks[indices]

# draw the bounding boxes and masks on the original image
for box, mask in zip(boxes, masks):
    # get the coordinates of the box
    x1, y1, x2, y2 = box.astype(int)
    # get the mask for the object
    mask = mask.squeeze()
    # resize the mask to match the original image size
    mask = cv2.resize(mask, (image.shape[1], image.shape[0]))
    # apply a threshold to binarize the mask
    mask = (mask > 0.5).astype(np.uint8)
    # create a color for the mask
    color = np.random.randint(0, 256, size=3)
    color = [int(c) for c in color]
    # draw the bounding box on the image
    cv2.rectangle(image, (x1, y1), (x2, y2), color, 2)
    # draw the mask on the image
    image[mask == 1] = cv2.addWeighted(image[mask == 1], 0.5, np.array(color), 0.5, 0)

# save or display the resulting image
cv2.imwrite("result.jpg", image)
cv2.imshow("Result", image)
cv2.waitKey(0)
cv2.destroyAllWindows()

报错信息 #

回溯（最近的调用最后）：
文件 "D:\MASTER_Project\testModel.py"，第27行，在中
output = model([image_tensor])
文件 "D:\MASTER_Project\venv\lib\site-packages\torch\nn\modules\module.py"，第1501行，在_call_impl中
return forward_call(*args, **kwargs)
文件 "D:\MASTER_Project\venv\lib\site-packages\torchvision\models\resnet.py"，第285行，在forward中
return self._forward_impl(x)
文件 "D:\MASTER_Project\venv\lib\site-packages\torchvision\models\resnet.py"，第268行，在_forward_impl中
x = self.conv1(x)
文件 "D:\MASTER_Project\venv\lib\site-packages\torch\nn\modules\module.py"，第1501行，在_call_impl中
return forward_call(*args, **kwargs)
文件 "D:\MASTER_Project\venv\lib\site-packages\torch\nn\modules\conv.py"，第463行，在forward中
return self._conv_forward(input, self.weight, self.bias)
文件 "D:\MASTER_Project\venv\lib\site-packages\torch\nn\modules\conv.py"，第459行，在_conv_forward中
return F.conv2d(input, weight, bias, self.stride,
TypeError: conv2d()接收到无效的参数组合 - 得到(list, Parameter, NoneType, tuple, tuple, tuple, int)，但预期为以下之一：

• (Tensor input, Tensor weight, Tensor bias, tuple of ints stride, tuple of ints padding, tuple of ints dilation, int groups)
  不匹配原因是部分参数类型无效：(!list of [Tensor]!, !Parameter!, !NoneType!, !tuple of (int, int)!, !tuple of (int, int)!, !tuple of (int, int)!, int)
• (Tensor input, Tensor weight, Tensor bias, tuple of ints stride, str padding, tuple of ints dilation, int groups)
  不匹配原因是部分参数类型无效：(!list of [Tensor]!, !Parameter!, !NoneType!, !tuple of (int, int)!, !tuple of (int, int)!, !tuple of (int, int)!, int)

解决方案 #

1. 修正输入参数类型 #

ResNet模型的forward方法接受的是Tensor类型输入，而非列表。你错误地将image_tensor用列表包裹传入，导致conv2d层接收到列表而非Tensor，触发类型错误。

将：

output = model([image_tensor])

改为：

# 先将tensor移到模型所在设备（CPU/GPU），同时转换为float32类型匹配模型要求
image_tensor = image_tensor.to(DEVICE).float()
output = model(image_tensor)

2. 额外注意事项 #

• 数据类型匹配：torch.from_numpy生成的是float64类型Tensor，而PyTorch模型默认使用float32，必须添加.float()转换类型，避免后续数据类型不匹配报错。
• 设备一致性：确保输入Tensor和模型在同一设备（CPU/GPU）上，否则会触发设备不匹配错误。你之前创建了data变量但未使用，直接使用未移设备的image_tensor会导致问题。
• 模型输出格式：普通ResNet是分类模型，输出的是类别概率张量，而非包含boxes、labels的字典。如果你的模型是基于ResNet的检测/分割模型（如Mask R-CNN），需单独确认输入格式，但当前报错的核心是输入类型错误，优先解决该问题。

修正后的完整关键代码片段 #

# preprocess the image
image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) # convert to RGB
image_resized = cv2.resize(image_rgb, (224, 224)) # resize to 224x224
image_normalized = image_resized / 255.0 # normalize to [0, 1]
image_tensor = torch.from_numpy(image_normalized) # convert to tensor
image_tensor = image_tensor.permute(2, 0, 1) # change the order of channels
image_tensor = image_tensor.unsqueeze(0) # add a batch dimension
# 修正：将tensor移到目标设备并转换为float32
image_tensor = image_tensor.to(DEVICE).float()

# pass the image tensor to the model and get the output
print(image)
output = model(image_tensor)

内容的提问来源于stack exchange，提问作者Tynoanswers