SegAnyGAussians icon indicating copy to clipboard operation
SegAnyGAussians copied to clipboard

The official implementation of SAGA (Segment Any 3D GAussians)

Results 80 SegAnyGAussians issues
Sort by recently updated
recently updated
newest added

您好!我选用nerf_llff_data中的fern数据集来训练3D高斯,预训练结束后得到了一个output文件夹: ![image](https://github.com/Jumpat/SegAnyGAussians/assets/145326984/b738074f-224b-4ee7-9338-8c1ac19c5ebb) 在接下来的过程中我输入指令:python train_contrastive_feature.py -m ./output/1a4fec7a-9/ 运行报错: ![image](https://github.com/Jumpat/SegAnyGAussians/assets/145326984/1fec0a59-c83d-4237-96c8-7dad4f300f1e) 想请问作者您是否也遇到过这种情况,应该如何解决呢?Any help is greatly appreciated!

when i ran extract_features.py using the command in github page, i was informed cuda out of memory. I'm new to coding and python project. After searching the Internet, I changed...

尊敬的作者您好!感谢您开源如此伟大的项目 我在预训练3D高斯的时候遇到了下面这个问题: ![image](https://github.com/Jumpat/SegAnyGAussians/assets/145326984/a4625b4f-b575-4278-8b6c-8e5b4646822c) 就是他这个training progress怎么只训练到23%就结束了?

作者,您好。您的SAGA让我受益匪浅。我这里有个问题想要问您,我该如何去通过SAGA分割得到自己想要的物体,就比如说我想从bicycle场景中分割得到自行车或者是自行车的轮胎。

抱歉再次打扰,上次询问您嘞SH2RGB的问题。近期我试了一下效果,但是转换结果和您论文中的有差距,请问是不是哪里处理不对? ![paper](https://github.com/Jumpat/SegAnyGAussians/assets/67455751/f7d2351b-2aa0-4b78-966d-039935d493b1) 对于这个场景,我读取了ply里的f_dc_0,f_dc_1,f_dc_2,将其传入了SH2RGB函数,结果作为rgb存入ply,最终用Viewer可视结果如下: ![viewer_room](https://github.com/Jumpat/SegAnyGAussians/assets/67455751/fc3572ec-8386-4a67-82d5-b3cce5cbe1ef) cloud_compare结果为: ![image](https://github.com/Jumpat/SegAnyGAussians/assets/67455751/97c59bd4-5934-499c-9fd3-c064deaa8d53)

When I use the original images in the dataset without processing, I will report an error like this root@I1993568c2b00601e90:/hy-tmp/SegAnyGAussians# python prompt_segmenting.py Looking for config file in ./output/e01280aa-a/cfg_args Config file found:...

hi,非常棒的工作,最近在学习代码。想问下一个报错怎么处理。 在运行prompt_segmenting.ipynb的postprocess_grad_based_statistical_filtering函数过程中遇到的。是函数调用错了吗?我看render调的是原始的gs渲染,所以没有mask这个参数。 感谢! `TypeError Traceback (most recent call last) Cell In[27], [line 7](vscode-notebook-cell:?execution_count=27&line=7) [4](vscode-notebook-cell:?execution_count=27&line=4) # write_ply('./segmentation_res/vanilla_seg.ply', selected_xyz) [6](vscode-notebook-cell:?execution_count=27&line=6) selected_xyz, thresh, mask_ = postprocess_statistical_filtering(pcd=selected_xyz.clone(), precomputed_mask = mask.clone(), max_time=1) ----> [7](vscode-notebook-cell:?execution_count=27&line=7) filtered_points,...

您好,您的SAGA项目让我受益匪浅! 我在阅读代码的过程中产生了一点疑惑:在使用 precomputed_mask 提取出整个场景中的目标场景对应的高斯球后,可以使用 render 渲染得到目标场景。但是这个过程中是怎么保持透明度的归一性的呢?因为一些背景中的高斯球没有被加入光栅化过程,会不会导致光栅化过程始终无法结束呢? 真诚盼望您的回答!

I am facing ![image](https://github.com/Jumpat/SegAnyGAussians/assets/56499208/fdf505a6-14ea-4d2e-8969-f00914d3a63a) I selected 3 points and set them to `input_point = np.array(selected_points)` ![image](https://github.com/Jumpat/SegAnyGAussians/assets/56499208/ea99b2b2-f05d-4103-8002-d65681e3b20a) I'm getting the SAM masks: ![image](https://github.com/Jumpat/SegAnyGAussians/assets/56499208/4560f430-11b3-4c64-adda-cb8e2a261712) I tried adding `.cuda()` (to `mask_pooling_prototype.unsqueeze(0)`) and removing...

您好,我在使用nerf_Synthetic数据集训练Gaussian时遇到了这个问题 RuntimeError: Function _RasterizeGaussiansBackward returned an invalid gradient at index 2 - got [0, 0, 3] but expected shape compatible with [0, 16, 3]。 我在Gaussian splatting的issues里找了一下没有具体解决措施,请问您有遇到过并解决吗?