NVlabs
Could FoundationPose detect multi objects with the same class(3D model) from one scene?
In the Nvidia isaac site, there was a demo about FoundationPose tracking multi objects with different classes(3D models) from one scene. https://nvidia-isaac-ros.github.io/repositories_and_packages/isaac_ros_pose_estimation/index.html
Could FoudationPose detect multi objects with the same class(3D model) from one scene? Some ones said in the papper, it seems FoudationPose could only detect one object from one scene. Is it correct?
I'm currently just creating a detector for every object, although that's a bit ugly
@liaozhangsheng Thank you for your reply. It looks great! Your bboxes are perfectly fit.
May I ask how you got your 3D models? Iphone app or others?
@liaozhangsheng Thank you for your reply. It looks great! Your bboxes are perfectly fit.
May I ask how you got your 3D models? Iphone app or others?
This is what I modeled myself in Blender. Perhaps other generated 3D models will also work, such as hunyuan-3d
Thank you so much for your information. Actually, I modeled in Blender too, but it did not work. Did you texture in Blender? May I ask what references you used?
Thank you so much for your information. Actually, I modeled in Blender too, but it did not work. Did you texture in Blender? May I ask what references you used?
I simply hand-drawn the texture
Thank you for your picture. Hand-drawn the texture sounds like a nice idea for Rubik's Cube. I taked the pictures and mapping them. Since there is only little information for FundationPose's modeling, and I am a novice in blender, I don't know what I did wrong.
@liaozhangsheng 您好，想问一下如果是一个场景中有多个相同物体（比如说您这张图中有多个魔方），因为物体完全相同，纹理也完全相同，直接用作者的foundationpose，导入这个物体的obj模型，是否就能得到想您这么好的位姿估计效果吗？需要去对分割模型做优化吗？
@liaozhangsheng 您好，想问一下如果是一个场景中有多个相同物体（比如说您这张图中有多个魔方），因为物体完全相同，纹理也完全相同，直接用作者的foundationpose，导入这个物体的obj模型，是否就能得到想您这么好的位姿估计效果吗？需要去对分割模型做优化吗？
随便一个分割模型都够用了，甚至直接用目标检测识别到的矩形框做掩码都行，估计效果还是很不错的，就是第一帧会很慢
@liaozhangsheng 您好，想问一下如果是一个场景中有多个相同物体（比如说您这张图中有多个魔方），因为物体完全相同，纹理也完全相同，直接用作者的foundationpose，导入这个物体的obj模型，是否就能得到想您这么好的位姿估计效果吗？需要去对分割模型做优化吗？
随便一个分割模型都够用了，甚至直接用目标检测识别到的矩形框做掩码都行，估计效果还是很不错的，就是第一帧会很慢
好滴，感谢您的回复，我试试～