Jiazhong Cen

你好，这个速度确实过于慢了，你能添加一些计时器来看看具体是哪一步比较慢吗？我们需要更多信息来debug。

我觉得不是数据集的问题，使用convert.py对数据集进行处理是没问题的，这个接口可能会有相机位姿坐标系对不齐之类的问题，但你可以试试。我依然推荐你对每个iteration各个步骤的时间开销进行一下诊断，包括数据读取步骤。

Hi, you can just remove the reliance on utils.contrastive_decoder_utils. This has been deprecated during our development.

Hi. What is your version of Scipy? I guess this is caused by the version mismatch. You can try to update it.

Oh, you should set the num_sampled_rays or the ray_sample_rate, which are set for reducing GPU memory consumption. ```sh python train_contrastive_feature.py -m output/73d7cf30-b --num_sampled_rays 1000 ``` I think this should work.

This is a challenging issue. As discussed in the appendix of our paper, SAGA does encounter multi-scale ambiguity when applied to open-vocabulary segmentation in complex scenes such as Lerf-OVS ramen....

你好，对这一问题的改善分三个层面： 第一，可以尝试在分割结果上再次进行分割（v2 branch saga gui中实现了这一功能） 第二，可以通过改进sam auto mask extraction的超参数来调整预提取的2D mask的分布，让这些2D mask能够涵盖待分割主题 第三，可以通过改进train contrastive feature时的损失函数来使affinity feature能够辨认出这一尺度上的物体 导致这一问题的可能性有很多，既可能是由于2D masks缺失对主体的描述，又可能是场景过于复杂，优化目标难以实现；也可能是场景尺度跨度过大。 由于main branch是旧版的分割方案，存在许多缺陷，我们推荐使用v2 branch的代码。另外如果还是无法解决问题，可以试试使用我们的另一个工作sa3d-gs。

Hello, we haven't test SAGA on the Treehill dataset. You must ensure that 3D-GS works well on it, which is the prerequisite for running SAGA on it. Moreover, more information...

Hi. You may need to check one of the boxes in Render option. The .pt file is a binary bitmap, whose entries indicate whether the corresponding 3D Gaussian belong to...

Hi, sorry for the late response. However, since SAGA is a per-scene optimization method and only requires one or two hours of training per scene, we did not keep the...