DataLoader reader thread raised an exception.

[FeiMa-REC]

报错如下：

文件相关配置：

【base.yml】

---

batch_size: 2 iters: 80000

train_dataset: type: Dataset dataset_root: ./ train_path: train_list.txt num_classes: 15 transforms:

• type: ResizeStepScaling min_scale_factor: 0.75 max_scale_factor: 2.0 scale_step_size: 0.25
• type: RandomPaddingCrop crop_size: [1024, 1024]
• type: RandomHorizontalFlip
• type: RandomRotation
• type: RandomDistort brightness_range: 0.4 contrast_range: 0.4 saturation_range: 0.4
• type: Normalize mode: train

val_dataset: type: Dataset dataset_root: ./ val_path: val_list.txt num_classes: 15 transforms:

• type: RandomHorizontalFlip
• type: Normalize mode: val

optimizer: type: sgd momentum: 0.9 weight_decay: 4.0e-4

learning_rate: value: 0.0075 decay: type: poly power: 0.9 end_lr: 0.0

loss: types:

• type: MixedLoss losses:
  • type: CrossEntropyLoss
  • type: LovaszSoftmaxLoss coef: [0.8, 0.2]
• type: MixedLoss losses:
  • type: CrossEntropyLoss
  • type: LovaszSoftmaxLoss coef: [0.8, 0.2] coef: [1, 0.4]

【train.yml】

---

base: '../base/SmartCar_base_Config.yml'

batch_size: 2 iters: 80000

model: type: DeepLabV3P backbone: type: ResNet101_vd output_stride: 8 multi_grid: [1, 2, 4] pretrained: https://bj.bcebos.com/paddleseg/dygraph/resnet101_vd_ssld.tar.gz num_classes: 15 backbone_indices: [0, 3] aspp_ratios: [1, 12, 24, 36] aspp_out_channels: 256 align_corners: False pretrained: null

【文件结构】：

---

【train_list】：

---

[BeyondYourself]

预训练模型直接用压缩文件？？ 都不解压么

[FeiMa-REC]

预训练模型直接用压缩文件？？ 都不解压么

执行训练流程的时候会在下载完预训练模型后自动解压吧？我看到有解压步骤的出现的。这个应该是dataloader的问题`(>﹏<)′

[michaelowenliu]

@YuBCS-Ma I guess data_root is wrong.

[FeiMa-REC]

@michaelowenliu I guess data_root is wrong.

My file tree is as follows :

&& My Config root is :

train_dataset: type: Dataset dataset_root: data/ train_path: data/train_list.txt

val_dataset: type: Dataset dataset_root: data/ val_path: data/val_list.txt

My train_list.txt :

But this time it still reported a mistake : <

[BeyondYourself]

你在配置文件中添加上transform 设置它的属性Resize试试 @YuBCS-Ma

[FeiMa-REC]

在配置文件中添加上transform 设置它的属性Resize试试 @BeyondYourself

啊，老哥，那个错误我解决了。resize没事，就是一直报数据读取的错。但是我的文件路径都对啊，不知道为什么跑不通。

刚刚跑的。

[michaelowenliu]

@YuBCS-Ma how did you run "train.py"? what is the absolute path of the data directory?

[FeiMa-REC]

@michaelowenliu how did you run "train.py"? what is the absolute path of the data directory?

In this way :

&& The absolute path of the data is : home/aistudio/data/

[BeyondYourself]

我之前遇到过报错和你一模一样的，结果是transform的属性值配置顺序有问题，resize最为最后属性值 @YuBCS-Ma

[FeiMa-REC]

我之前遇到过报错和你一模一样的，结果是transform的属性值配置顺序有问题，resize最为最后属性值 @BeyondYourself

这是我的Transform

[BeyondYourself]

你先试试属性值只加上Resize和Normalize @YuBCS-Ma

[FeiMa-REC]

你先试试属性值只加上Resize和Normalize @BeyondYourself

我试了只加Normalize，还是一样的报错，应该不是这个的问题。

[BeyondYourself]

这种报错有的时候真的不太好找，建议debug看看

[michaelowenliu]

@YuBCS-Ma Hi, please try running train.py in PaddleSeg-Release-v2.0 directory, and set data_root to home/aistudio/data/, set train_path: home/aistudio/data/train_list.txt.

[BeyondYourself]

@michaelowenliu 所以在读取文件是不是应该加一个验证文件以及路径是否存在

[michaelowenliu]

@BeyondYourself Hi, a great idea! Hope you can help to add this feature and submit the PR.

[BeyondYourself]

OK，我试试看

[wuyefeilin]

我之前遇到过报错和你一模一样的，结果是transform的属性值配置顺序有问题，resize最为最后属性值 @BeyondYourself

这是我的Transform

RandomRotation 会改变图片大小，把它放到RandomPaddingCrop之前

[BeyondYourself]

@wuyefeilin 出个transform的配置宝典吧

[wuyefeilin]

@wuyefeilin 出个transform的配置宝典吧

感谢反馈，我们会充分考虑你的建议

[FeiMa-REC]

@michaelowenliu @wuyefeilin

Finally, it works！！！：）

I should set_env PYTHONPATH='/home/aistudio/PaddleSeg-release-v2.0/' and change dataset_root , train_path, also should put the RandomRotation before the RandomPaddingCrop.

Thx! ! !

[Exusia-0]

我之前遇到过报错和你一模一样的，结果是transform的属性值配置顺序有问题，resize最最后的属性值@BeyondYourself

这是我的Transform

RandomRotation会改变图片大小，把它放到RandomPaddingCrop之前

确实是这个问题