liuyuan-pal / Gen6D

[ECCV2022] Gen6D: Generalizable Model-Free 6-DoF Object Pose Estimation from RGB Images

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

输出结果视频发生翻转,效果不如预期(Video results flipped)

coolman-ai opened this issue · comments

[我在ubuntu系统下测试gen6d模型时,按照作者所给权重参数直接测试(测试视频为作者提供的视频,点云模型也使用作者提供的原始模型),但是输出的结果视频被翻转,且结果似乎不如预期,如链接所示,请问有人遇到相似的情况吗?麻烦解答,十分感谢!]
When I tested the gen6d model under the ubuntu system, I directly tested it according to the weight parameters given by the author (the test video was provided by the author. The point cloud model also used the original model provided by the author), but the output video was flipped and the results seemed to be not as expected, as shown in the link. Has anyone encountered a similar situation? Could you please answer? Thank you very much!
https://github.com/liuyuan-pal/Gen6D/assets/82800272/57c39332-bdbe-4b3e-9d54-02c7d4b32ec4

hi,你可以check一下那个把video拆分成图像的那个步骤,它有时候和opencv版本有关,图像是否会翻转。

我也遇到了相似情况,我的解决措施是执行python predict.py 代码时将--transpose删除,就可以解决翻转问题。不过我的生成结果和你的类似,我不明白的一个问题是预测生成的框明显比物体大比较多,而且不太稳定。目前我并不知道怎么解决这个问题