https://longtimenohack.com/Jianfei GuoHugo -- gohugo.iozh-CNTue, 21 Nov 2023 14:44:27 +0800https://longtimenohack.com/posts/paper_reading/2023_cui_alethnerf/Tue, 21 Nov 2023 14:44:27 +0800https://longtimenohack.com/posts/paper_reading/2023_cui_alethnerf/编者按 主要任务：利用低光照的图，重建原始的充足光照的场景 主要思想：在nerf可微渲染时，从物体到视点，沿光线方向衰减亮度，把在整个空间中的光https://longtimenohack.com/posts/datasets/waymo_asset_v1v2/Thu, 09 Nov 2023 15:22:05 +0800https://longtimenohack.com/posts/datasets/waymo_asset_v1v2/preliminaries: waymo v1 标注检查 lidar_camera_projection 标注 range_images, camera_projections, seg_labels, range_image_top_pose = frame_utils.parse_range_image_and_camera_projection(frame) points, cp_points = frame_utils.convert_range_image_to_point_cloud( frame, range_images, camera_projections, range_image_top_pose, ri_index=0) # first return points 是一个列表，每个 item 来自不同的 laser；每个 item 是从 range_images 计算出的 [N, 3] 尺寸点云，N 是当前https://longtimenohack.com/posts/basics_maths/permutohedral_lattice/Mon, 11 Sep 2023 09:44:43 +0800https://longtimenohack.com/posts/basics_maths/permutohedral_lattice/Background maths Permutation: 置换（排列） 定义：既是一种排列，也是一种映射 n 个元素不缺失不重复的不同排列方式；每一种叫做一种置换 置换也可以定义成一种集合到自身的映射https://longtimenohack.com/posts/paper_reading/2023cvpr_rosu_permutosdf/Wed, 06 Sep 2023 19:00:00 +0800https://longtimenohack.com/posts/paper_reading/2023cvpr_rosu_permutosdf/编者语：主要是把 NGP 的思想推广到了一种对高维输入更友好的格点 (lattice) 上 背景：permutohedral lattice (排列多面体点阵) 背景数学知识详见我的 这篇笔https://longtimenohack.com/posts/paper_reading/2022siggraph_mueller_instant/Wed, 23 Feb 2022 19:00:00 +0800https://longtimenohack.com/posts/paper_reading/2022siggraph_mueller_instant/代码框架阅读 (包含详细梯度传导过程) 点击图片链接跳转至 processon instant-ngp 代码梳理 Tiny-cuda-nn 我为 tiny-cuda-nn 实现的 encoding 二阶导 PR链接https://longtimenohack.com/posts/basics_maths/deformation/Tue, 27 Jul 2021 19:44:43 +0800https://longtimenohack.com/posts/basics_maths/deformation/resource MIT Continuum Mechanics 连续介质力学 课程笔记，Chapter 2 重要参考资料：非常详尽、非常严谨、非常简洁明了，与线代知识紧密接续 虽然属于连续介质力学中的内容https://longtimenohack.com/posts/nerf/nerf_on_surface/Sat, 17 Jul 2021 19:44:43 +0800https://longtimenohack.com/posts/nerf/nerf_on_surface/https://longtimenohack.com/posts/paper_reading/2021_chen_geosim/Tue, 13 Jul 2021 19:44:43 +0800https://longtimenohack.com/posts/paper_reading/2021_chen_geosim/编者按 动态部件细节：轮子转动时没有的；但是这个可能对自动驾驶感知并无影响？ 做的事情都是把物体添加到已有观测中； 并没有新的整体观测视角； 并没有https://longtimenohack.com/posts/paper_reading/2021_martin_nerfw/Tue, 13 Jul 2021 19:44:43 +0800https://longtimenohack.com/posts/paper_reading/2021_martin_nerfw/编者按 Motivation 效果：从网络旅游照片集中重建一个三维物体 diss之前 NeRF：cons：需要几何、材质、光学都是静态的假设； task：做了什么 本文想https://longtimenohack.com/posts/paper_reading/2021_ost_neural_scene_graph/Tue, 13 Jul 2021 19:44:43 +0800https://longtimenohack.com/posts/paper_reading/2021_ost_neural_scene_graph/编者按 不是传统的 scene graph 中的 graph 的定义，事实上较为 hand-crafted； 额外把类别级表征模型也放到图里了，并且额外定义了一具体物体表征到类别级