train.md要求准备的深度图是png格式，但是我运行train.py后却显示因为png格式没有"near""far"出错了

感谢开源，作者你好~
我下载的TUM数据集，后按照训练要求做好了数据集的格式，但是在运行train.py之后，却因为读不到深度图的“near”和“far”属性，导致读取深度图失败了，请问这是我数据集来源的问题吗？
另外我还想请问一下，我的meta.json文件中，内参矩阵中带负数，对于运行代码有影响吗？因为我在运行train的dataloader的板块

6.1 The image and depth are resized first to approximately the same pixel size as the target image with PIL's antialiasing resampling下的代码

image = np.array(Image.fromarray(image).resize((rescaled_w, rescaled_h), Image.Resampling.LANCZOS))的时候，它如此提示
Exception has occurred: ValueError
height and width must be > 0

望您回复，谢谢。若我提问并不清楚，也请您联系我。

Start with train.py’s dataloader section, especially the PIL resize call shown in the report, and inspect how the TUM images, depth PNG metadata, and meta.json intrinsics are read. Reproduce the failure with the reported dataset and verify why the resized height or width becomes non-positive; done means the dataset loads through training with valid dimensions.