Gen6D for AR application?

[Kahina-88]

Would you provide me with tips on how to augment a virtual object based on Gen6D pose estimation? is using unity3D necessary? Where are the 3D bounding box coordinates registered?

[liuyuan-pal]

Hi, I use blender to render the object. You just export the poses and render the object in blender with the pose and then composite the rendered object with the original query image.

[AmokraneIlhem]

@liuyuan-pal
Please, how to export the poses? thank you in advance.

[Kahina-88]

Would you provide me with tips to export the poses? How to relate Blender with Geb6D?

[liuyuan-pal]

https://github.com/liuyuan-pal/Gen6D/blob/f95e82b075a5a61f9ef436c7756f4b07cffd172a/predict.py#L69

This line predicts pose. However, Blender uses the coordinate system of opengl, which is different from pose here and it sometimes depends on the model you used for rendering. You may refer to the script in pvnet-render https://github.com/zju3dv/pvnet-rendering

[lucker26]

https://github.com/liuyuan-pal/Gen6D/blob/f95e82b075a5a61f9ef436c7756f4b07cffd172a/predict.py#L69

这条线预测姿势。但是，Blender 使用的是 opengl 的坐标系，这与这里的姿势不同，有时取决于您用于渲染的模型。您可以参考 pvnet-render 中的脚本 https://github.com/zju3dv/pvnet-rendering

请问np.save(f'{str(output_dir)}/images_out/{que_id}-pose.npy', pose_pr)这里保存的姿态posepr和pose = pnp（object_bbox3d， pts， K) ，这两者都是表示物体的姿态吗？两者有什么区别！