代码

256×256 --> 224×224

import torch
import torch.nn as nn# 假设 x 是你的特征图，形状为 (4, 32, 256, 256)
x = torch.randn(4, 32, 256, 256)# 方法一：使用自适应平均池化调整大小
adaptive_avg_pool = nn.AdaptiveAvgPool2d((224, 224))
x_pooled_avg = adaptive_avg_pool(x)print(x_pooled_avg.shape)  # 输出形状应该是 (4, 32, 224, 224)# 方法二：使用自适应最大池化调整大小
adaptive_max_pool = nn.AdaptiveMaxPool2d((224, 224))
x_pooled_max = adaptive_max_pool(x)print(x_pooled_max.shape)  # 输出形状应该是 (4, 32, 224, 224)

224×224 --> 256×256

import torch
import torch.nn as nn# 创建一个随机的特征图，形状为 (4, 32, 224, 224)
feature_map = torch.randn(4, 32, 224, 224)# 定义双线性插值的上采样层
upsample = nn.Upsample(size=(256, 256), mode='bilinear', align_corners=False)# 应用上采样
upsampled_feature_map = upsample(feature_map)
print(upsampled_feature_map.shape)  # 输出应为 (4, 32, 256, 256)