CGI-Stereo icon indicating copy to clipboard operation
CGI-Stereo copied to clipboard

Published 20 hours ago • verified publisher icon gangweiX
→

Metadata

A novel neural network architecture that can concurrently achieve real-time performance, competitive accuracy, and strong generalization ability.

Results 8 CGI-Stereo issues
Sort by recently updated
recently updated
newest added

关于验证问题

怎么将左右两张图片作为输入，输出视差图。

MR-STUZHANG avatar MR-STUZHANG

感谢您的贡献，请问按论文所说提供的Senceflow预训练模型是20epoch还是，20 + 20微调之后的呢

tangqi1012 avatar tangqi1012

关于KITTI数据集测试的问题

6 comment

你好，我看见代码中训练的默认设置是使用KT2015和KT2012的所有数据进行训练，但是把KT15的一部分用来做验证。这样验证集被训练集包括了，这样验证是合理的吗

chenzhaojie1997 avatar chenzhaojie1997

您好，想请问您论文中关于泛化性的问题

1 comment

首先就是真实值是如何生成图片的呢 ![image](https://github.com/gangweiX/CGI-Stereo/assets/115446508/b578b491-7759-4572-9323-cf57fa99dfe3) 然后，您论文中middlebury的bad2.0为13.5，摩托车生成的很漂亮，然而我测下来bad2.0为11.8，视差颜色却是很深呢 ![image](https://github.com/gangweiX/CGI-Stereo/assets/115446508/88a9b628-afae-46cb-9dfa-501bd83fa252) ![image](https://github.com/gangweiX/CGI-Stereo/assets/115446508/f96cac3f-04f1-4585-8a3d-5fc324a7e242) ![image](https://github.com/gangweiX/CGI-Stereo/assets/115446508/e1267ba1-4346-42e8-a901-1d6861ca2bc2)

1434205229 avatar 1434205229

D1_metric问题

![image](https://github.com/gangweiX/CGI-Stereo/assets/49083219/7db70127-7311-4a0c-be40-645047e71a17) 您好，请问这个的D1指标的阈值是不是有点问题？如果我没理解错这里3的单位是像素，而标签和推理结果都是归一化的值。不知道我理解的是否正确？

mzcxiaohuanxiong avatar mzcxiaohuanxiong

请问合并kitti微调时要想找出kitti2012上最好的模型，需要改为kitti12_Val吗

1 comment

![image](https://github.com/gangweiX/CGI-Stereo/assets/115446508/2bc8a92d-0d24-41c6-bdd0-9ec5de234459)

1434205229 avatar 1434205229

你好，请问可以提供Middlebury的权重进行测试吗？

Hlll-huang avatar Hlll-huang

前辈您好，请问这篇文章被接收了吗，什么期刊呀，谢谢！

flyxuexi avatar flyxuexi

← Metadata

290
Stars
29
Forks
Watchers

Owner

verified publisher icon gangweiX

Metadata

A novel neural network architecture that can concurrently achieve real-time performance, competitive accuracy, and strong generalization ability.

Back