版面分析模型加载很多警告

[Kyo1234567]

🔎 Search before asking

• I have searched the PaddleOCR Docs and found no similar bug report.
• I have searched the PaddleOCR Issues and found no similar bug report.
• I have searched the PaddleOCR Discussions and found no similar bug report.

🐛 Bug (问题描述)

我在训练PaddleDetection的版面分析，使用的配置文件是picodet_lcnet_cdla_layout，遇到很多警告，似乎模型和配置不匹配：
[03/04 10:00:48] ppdet.utils.checkpoint INFO: ['fc.bias', 'fc.weight', 'last_conv.weight'] in pretrained weight is not used in the model, and its will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn._mean is unmatched with the shape [64] in model neck.bottom_up_blocks.0.blocks.0.conv1.bn._mean. And the weight conv1.bn._mean will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn._variance is unmatched with the shape [64] in model neck.bottom_up_blocks.0.blocks.0.conv1.bn._variance. And the weight conv1.bn._variance will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn.bias is unmatched with the shape [64] in model neck.bottom_up_blocks.0.blocks.0.conv1.bn.bias. And the weight conv1.bn.bias will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn.weight is unmatched with the shape [64] in model neck.bottom_up_blocks.0.blocks.0.conv1.bn.weight. And the weight conv1.bn.weight will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16, 3, 3, 3] in pretrained weight conv1.conv.weight is unmatched with the shape [64, 64, 1, 1] in model neck.bottom_up_blocks.0.blocks.0.conv1.conv.weight. And the weight conv1.conv.weight will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn._mean is unmatched with the shape [64] in model neck.bottom_up_blocks.1.blocks.0.conv1.bn._mean. And the weight conv1.bn._mean will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn._variance is unmatched with the shape [64] in model neck.bottom_up_blocks.1.blocks.0.conv1.bn._variance. And the weight conv1.bn._variance will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn.bias is unmatched with the shape [64] in model neck.bottom_up_blocks.1.blocks.0.conv1.bn.bias. And the weight conv1.bn.bias will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn.weight is unmatched with the shape [64] in model neck.bottom_up_blocks.1.blocks.0.conv1.bn.weight. And the weight conv1.bn.weight will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16, 3, 3, 3] in pretrained weight conv1.conv.weight is unmatched with the shape [64, 64, 1, 1] in model neck.bottom_up_blocks.1.blocks.0.conv1.conv.weight. And the weight conv1.conv.weight will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn._mean is unmatched with the shape [64] in model neck.top_down_blocks.0.blocks.0.conv1.bn._mean. And the weight conv1.bn._mean will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn._variance is unmatched with the shape [64] in model neck.top_down_blocks.0.blocks.0.conv1.bn._variance. And the weight conv1.bn._variance will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn.bias is unmatched with the shape [64] in model neck.top_down_blocks.0.blocks.0.conv1.bn.bias. And the weight conv1.bn.bias will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn.weight is unmatched with the shape [64] in model neck.top_down_blocks.0.blocks.0.conv1.bn.weight. And the weight conv1.bn.weight will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16, 3, 3, 3] in pretrained weight conv1.conv.weight is unmatched with the shape [64, 64, 1, 1] in model neck.top_down_blocks.0.blocks.0.conv1.conv.weight. And the weight conv1.conv.weight will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn._mean is unmatched with the shape [64] in model neck.top_down_blocks.1.blocks.0.conv1.bn._mean. And the weight conv1.bn._mean will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn._variance is unmatched with the shape [64] in model neck.top_down_blocks.1.blocks.0.conv1.bn._variance. And the weight conv1.bn._variance will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn.bias is unmatched with the shape [64] in model neck.top_down_blocks.1.blocks.0.conv1.bn.bias. And the weight conv1.bn.bias will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16] in pretrained weight conv1.bn.weight is unmatched with the shape [64] in model neck.top_down_blocks.1.blocks.0.conv1.bn.weight. And the weight conv1.bn.weight will not be loaded
[03/04 10:00:48] ppdet.utils.checkpoint INFO: The shape [16, 3, 3, 3] in pretrained weight conv1.conv.weight is unmatched with the shape [64, 64, 1, 1] in model neck.top_down_blocks.1.blocks.0.conv1.conv.weight. And the weight conv1.conv.weight will not be loaded

🏃‍♂️ Environment (运行环境)

windows 10, paddlepaddle-gpu 2.6.1

🌰 Minimal Reproducible Example (最小可复现问题的Demo)

无

[GreatV]

看你的日志，警告信息表明预训练权重文件中的某些参数未被加载，或者由于形状不匹配导致某些参数无法正确映射到模型结构。这通常发生在以下几种情况下：

可能的原因

1. 模型结构与预训练权重不匹配

   • 你可能修改了 picodet_lcnet_cdla_layout 配置文件的某些参数，如 Neck 或 Backbone 的结构，导致与预训练模型的参数形状不同。
   • 例如，从日志可见 conv1.bn._mean 在预训练权重中是 [16]，但在当前模型中是 [64]，说明 conv1 相关的通道数发生了变化。

2. 预训练模型与当前使用的 PaddleDetection 版本不完全兼容

   • 你使用的 paddlepaddle-gpu 2.6.1 可能与该预训练模型的版本不匹配。如果预训练模型是基于 PaddleDetection 更新后的结构训练的，而你的环境版本较老，就可能导致加载问题。

3. 不同的 pretrained_model 文件

   • pretrained_model 可能指定了错误的路径，或者该路径下的模型权重文件不完整或损坏，导致部分权重无法加载。

---

解决方案

1. 确保 pretrained_model 路径正确

在 picodet_lcnet_cdla_layout.yml 配置文件或 CLI 传参中正确指定预训练权重:

architecture: Picodet
pretrained_model: path/to/the/correct/pretrained_model

2. 检查 PaddleDetection 版本是否匹配

使用以下命令更新 PaddleDetection 并重新尝试：

python -m pip install --upgrade paddlepaddle-gpu -i https://pypi.tuna.tsinghua.edu.cn/simple

3. 确保 picodet_lcnet_cdla_layout.yml 配置文件的 Neck 与预训练模型匹配

• 你可以参考官方 PaddleDetection 提供的模型结构，确认 Neck（尤其是 bottom_up_blocks 和 top_down_blocks）与预训练模型结构一致：
• 若你修改过 yaml 配置文件（如 num_filters 或 conv 层的参数），请尝试改回默认值。

4. 额外尝试：重新下载预训练模型

如果仍然存在问题，可以尝试重新下载预训练模型，确保权重文件完整：

wget https://path-to-correct-pretrained-model -O pretrained_model.pdparams

---

总结

你的问题主要是模型结构变动导致部分权重无法匹配，可以按照 检查配置、更新 PaddleDetection、重新下载权重 这几个步骤进行调试，优先确保模型结构和预训练权重是匹配的。

如果问题仍然存在，可以提供 picodet_lcnet_cdla_layout.yml 的详细配置内容，以便进一步分析。🚀

---

Response generated by 🤖 feifei-bot | chatgpt-4o-latest

⚠️ This is an AI-generated response. Please verify any critical information.