关于test_simple_SQL_config.py可视化

[chaoying0115]

您好，非常感谢您的出色工作。我用test_simple_SQL_config.py进行图片测试，得到了以下图片，与论文风格有些不一致，请问是哪里出了问题？

[Lavreniuk]

@chaoying0115 , thank you for interesting in our model, have you set checkpoints path?
please find attached image and result for which I just double checke


d.

[chaoying0115]

您好 非常感谢您的回复。我是SQLdepth那边下载的模型进行测试，但测出的情况如下
/

这是代码的运行情况

不知道问题出在哪里 期待得到您的回复！

[Lavreniuk]

for locating the problem, could you run the inference on the image that I provided?
if you will get same result as I on that image, it means everything work, if not - so something wrong with your set up.

[chaoying0115]

您好 我是用您提供的图像进行推理 得到以下结果

这是运行截图和相关文件设置

args_test_simple_kitti_320x1024.txt
我使用的是SOLdepth 提供的权重 https://drive.google.com/file/d/1_BHfGXqVsE4RrCM58_5nEzQHJmyxDsQO/view

[Lavreniuk]

ok, but as I saw you used other config than I provided in the readme file.
Please use this one and just download the pretrained weights and change the path for them.
conf/cvnXt.txt

[chaoying0115]

好的 谢谢您。我按照您的文件，下载了https://drive.google.com/file/d/14Hb8UsjraLWk1TvtMPd5moEDMlDxiHkC/view 的权重
文件设置和运行结果如下，似乎出现了权重不匹配的问题 是我权重下载错了吗？

[Lavreniuk]

I think you have downloaded some other model.
Please double check that you have downloaded from the link that I provided in readme:
https://huggingface.co/MykolaL/SPIdepth/tree/main/kitti

[chaoying0115]

谢谢 在您指导下我得到与您一样的可视化。我发现我先前的错误是发生在自监督Resnet-50预训练模型的加载上，他们似乎不会去加载文件提供相关权重文件，如图所示，而您提供的是全监督情况下的，可以加载文件提供的权重，所以没有问题

您有尝试做过自监督模型的可视化吗？能够得到正确风格的可视化图像？

[Lavreniuk]

yes, you could do same for unsupervised model, but you need to train it by yourself, because I provided only final weights.
And if you find the project helpful, starring the repo would be appreciated.