Jiazhong Cen
Jiazhong Cen
This: --config=configs/llff/trex.py It should be seg_trex.py also. During segmentation you must switch to the new config for building the mask grids.
The lego bulldozer is from MIP360-kitchen scene not the blender. We have provided it [here](https://github.com/Jumpat/SegmentAnythingin3D/blob/main/configs/nerf_unbounded/kitchen.py).
Hi, this is caused by wrongly installed cuda. Can you try torch.cuda.is_available() in this environment? What is the version of the installed CUDA.
They are two different projects. You can refer to the papers of [SA3D](https://arxiv.org/abs/2304.12308) and [SAGA](https://arxiv.org/abs/2312.00860) for details.
This is not an easy issue. The nerf studio branch is maintained by my collaborator. The 3D-GS version currently does not support text prompts. The text-prompt based segmentation is implemented...
分割好后的三维模型实际上就是以320**3的grid形式存储的，请参考lib/grids.py部分的代码。
分割好的结果是以网格形式保存的，只需要把DenseGrid中对应的变量grid取出来，但具体要导出成什么格式需要根据下游的需求自己修改并存储。
seg_dvgo和seg_dcvgo中的seg_mask_grid
是的，是一个tensor，可以转换成np. array
> 我用的nerfstudio 版本 调用ns-export 导出点云时候出现 pipeline.datamanager.train_pixel_sampler.num_rays_per_batch = self.num_rays_per_batch AttributeError: 'NoneType' object has no attribute 'num_rays_per_batch' 不知道如何修改导出mask的点云 你好，我们并没有在nerfstudio上严格实现点云导出等功能，这可能导致无法预料的错误，目前我们没有维护该分支的计划，你可能需要根据nerfstudio的源码对这部分进行调整。