更换BirdRecognizerPython模型.pb后输入维度不匹配问题求助
图像分类模型维度不匹配问题解决建议
问题描述
更换model.pb文件后,运行图像分类代码出现维度不匹配错误:
File "/Users/Desktop/poc/BirdRecognizerPython/bird_recog.py", line 60, in <module> predictions = sess.run(prob_tensor, {input_node: [augmented_image] }) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/Users/Documents/virtualenv_dir/ka_ai/lib/python3.11/site-packages/tensorflow/python/client/session.py", line 972, in run result = self._run(None, fetches, feed_dict, options_ptr, ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/Users/Documents/virtualenv_dir/ka_ai/lib/python3.11/site-packages/tensorflow/python/client/session.py", line 1189, in _run raise ValueError( ValueError: Cannot feed value of shape (1, 227, 227, 3) for Tensor Placeholder:0, which has shape (None, 224, 224, 3)
尝试添加以下代码后,出现新的维度错误:
augmented_image = np.expand_dims(augmented_image, axis=0) # Convert to float32 augmented_image = augmented_image.astype(np.float32)
错误信息:
ValueError: Cannot feed value of shape (1, 1, 227, 227, 3) for Tensor Placeholder:0, which has shape (None, 224, 224, 3)
核心问题分析
- 新模型的输入占位符要求尺寸为
(None, 224, 224, 3),但当前输入图像尺寸是227x227,导致维度不匹配。 - 原代码中
{input_node: [augmented_image] }已经将单张图像包装成了(1, 227, 227, 3)的批量维度,额外添加的np.expand_dims会多增加一个维度,变成(1,1,227,227,3),进一步加剧维度错误。
解决步骤
调整图像尺寸至模型要求的224x224
在图像预处理阶段,将augmented_image的尺寸修改为224x224。例如使用OpenCV调整:import cv2 # 假设augmented_image是(227,227,3)的数组 augmented_image = cv2.resize(augmented_image, (224, 224))或使用PIL库:
from PIL import Image import numpy as np augmented_image = np.array(Image.fromarray(augmented_image).resize((224, 224)))移除多余的维度扩展代码
删除你添加的augmented_image = np.expand_dims(augmented_image, axis=0)这一行,保留类型转换(如果模型需要float32类型输入):augmented_image = augmented_image.astype(np.float32)验证输入维度
调整后可以打印维度确认:print(augmented_image.shape) # 应输出(224,224,3) print(np.array([augmented_image]).shape) # 应输出(1,224,224,3),与模型占位符匹配
内容的提问来源于stack exchange,提问作者datagekk
相关产品推荐
相关产品推荐

