Yuan Liu

https://github.com/liuyuan-pal/Gen6D/blob/50aa71b2dbb3198772c025d54d0dd191a918982c/prepare.py#L28 有可能你是要flip一下

另外gen6d在这种clutter背景（前景背景颜色差不多），然后存在一定遮挡的情况下，可能是比较容易出错，没有那么robust。这里没有针对occlusion有任何设计哈

那就是你的z方向选错了，meta那个

因为那个fitting的normal可能是正的也可能是反的，你可以看看normal的图像，决定哪个方向是对的~

你可以看看predict.py，这里其实涉及到三个网络，可能你得分别处理三个。

Hi! Sorry for the late reply. I've checked the evaluation codes and I found I forgot to add a "pair" directory (downloaded from the OANet repo) in the "data" directory....

Hi, thanks for your interest! All parameters named by "K" are intrinsic parameters. R is the rotation and T is the translation from the extrinsic matrix (i.e. camera pose).

There should be known intrinsic parameters on MegaDepth dataset also so you may directly use them in your cases. As you can see, extend to a new dataset requires writing...

1. Yes, you can use the provided intrinsics. 2. If you would like to use the eval.py script you should create a new subclass. 3. filter_cur_name is just used in...

I'll release the model this weekend.