wuvin
Always take risks!
Toggle navigation
wuvin
主页
实验室的搬砖生活
机器学习
公开的学术内容
公开的其他内容
About Me
归档
标签
友情链接
ZYQN
ihopenot
enigma_aw
hzwer
杨宗翰
ECCV2022 NeRF 相关
2022-11-23 16:11:33
198
0
0
wuvin
* 参考了 https://markboss.me/post/nerf_at_eccv22/ ### Streamable Neural Fields * ![title](https://leanote.com/api/file/getImage?fileId=637ddb09ab64411710a660aa) * ![title](https://leanote.com/api/file/getImage?fileId=637ddb2cab64411710a660ab) * ![title](https://leanote.com/api/file/getImage?fileId=637ddb46ab64411709a80dab) * ![title](https://leanote.com/api/file/getImage?fileId=637dde56ab64411709a80dc8) * 有意思的工作,可以用于 NeRF 的模糊图预览。 ### Conditional-Flow NeRF * ![title](https://leanote.com/api/file/getImage?fileId=637ddf22ab64411710a660c5) * ![title](https://leanote.com/api/file/getImage?fileId=637ddf2fab64411709a80dcd) * ![title](https://leanote.com/api/file/getImage?fileId=637ddf3dab64411709a80dcf) * 估计每一部分重建的 Uncertanty. ### R2L * ![title](https://leanote.com/api/file/getImage?fileId=637de2a3ab64411709a80dfa) * ![title](https://leanote.com/api/file/getImage?fileId=637de2cdab64411709a80dfd) * 把 NeRF 转成 LightFeild 加速渲染。 ### AdaNeRF * 降低采样数量来加速渲染。 * ![title](https://leanote.com/api/file/getImage?fileId=637de378ab64411709a80e04) ### Intrinsic Neural Fields * NeualTexture 的竞争对手,提出了一种新的Basis。 * ![title](https://leanote.com/api/file/getImage?fileId=637de529ab64411710a660f2) * ![title](https://leanote.com/api/file/getImage?fileId=637de4a4ab64411710a660ef) ### Beyond Periodicity * 测试了不同的NeRF 的激活函数 * ![title](https://leanote.com/api/file/getImage?fileId=637df4cfab64411710a66163) * ![title](https://leanote.com/api/file/getImage?fileId=637df4edab64411709a80e84) * ![title](https://leanote.com/api/file/getImage?fileId=637df50bab64411709a80e86) * ![title](https://leanote.com/api/file/getImage?fileId=637df61bab64411709a80e8f) ### NeDDF * 优化具有不确定边界的物体(烟雾、毛球、玻璃等)的距离场 * ![title](https://leanote.com/api/file/getImage?fileId=637e02f0ab64411709a80f00) ### BungeeNeRF: Progressive Neural Radiance Field for Extreme Multi-scale Scene Rendering * 咋这么眼熟呢,这不是 city NeRF 的图? * 查了一下是 CityNeRF 改名重投了。 * ![title](https://leanote.com/api/file/getImage?fileId=637e0246ab64411710a661e4) * ![title](https://leanote.com/api/file/getImage?fileId=637e0296ab64411709a80efd) ### SNeS: Learning Probably Symmetric Neural Surfaces from Incomplete Data * 根据物体本身的对称性增加的正则化。但这篇文章一共就做了一个数据集??? * ![title](https://leanote.com/api/file/getImage?fileId=637e03a7ab64411709a80f03) * ![title](https://leanote.com/api/file/getImage?fileId=637e044aab64411710a66202) * ![title](https://leanote.com/api/file/getImage?fileId=637e0479ab64411709a80f07) ### HDR-Plenoxels: Self-Calibrating High Dynamic Range Radiance Fields * 多个 LDR 图像合成 HDR 图像 * ![title](https://leanote.com/api/file/getImage?fileId=637e04d3ab64411710a6620c) ### MINER: Multiscale Implicit Neural Representation * 多个 Scale 的MLP * ![title](https://leanote.com/api/file/getImage?fileId=637e0556ab64411710a66211) * ![title](https://leanote.com/api/file/getImage?fileId=637e0537ab64411710a66210) * ![title](https://leanote.com/api/file/getImage?fileId=637e05c7ab64411710a66214) ### ShAPO: Implicit Representations for Multi-Object Shape, Appearance, and Pose Optimization * 尝试单RGBD图像复原三维。但需要先预定义好一些三维形状。 * ![title](https://leanote.com/api/file/getImage?fileId=637e0662ab64411710a66218) * ![title](https://leanote.com/api/file/getImage?fileId=637e07b9ab64411710a66225) ### KeypointNeRF: Generalizing Image-based Volumetric Avatars using Relative Spatial Encoding of Keypoints * 关键点定位+人体先验 = few-shot 的人体重建 * ![title](https://leanote.com/api/file/getImage?fileId=637e08b5ab64411709a80f27) * ![title](https://leanote.com/api/file/getImage?fileId=637e0928ab64411710a66232) ### Learning Dynamic Facial Radiance Fields for Few-Shot Talking Head Synthesis * few-shot 的视频人脸重建,其中NeRF is conditioned on 2D appearance and audio features,怎么感觉适合拿来干坏事 * ![title](https://leanote.com/api/file/getImage?fileId=637e09d6ab64411710a66237) * ![title](https://leanote.com/api/file/getImage?fileId=637e09e5ab64411710a66238) ### Semantic-Aware Implicit Neural Audio-Driven Video Portrait Generation * NeRF Deformation + audio = 语音生成人脸移动和面部变化序列 * ![title](https://leanote.com/api/file/getImage?fileId=637e4ec2ab64411709a81131) ### SparseNeuS: Fast Generalizable Neural Surface Reconstruction from Sparse Views * few-shot NeuS, 先用一些 SDF 先验 * 里面用到一个 Patch-based color blending 还比较有意思 * ![title](https://leanote.com/api/file/getImage?fileId=637e0a54ab64411710a6623b) * ![title](https://leanote.com/api/file/getImage?fileId=637e0a68ab64411709a80f37) * ![title](https://leanote.com/api/file/getImage?fileId=637e0aa7ab64411709a80f3a) ### NeuMesh: Learning Disentangled Neural Mesh-based Implicit Field for Geometry and Texture Editing * 最大的问题是,这个工作需要预先输入一个mesh。 * ![title](https://leanote.com/api/file/getImage?fileId=637e393eab64411710a663c3) * ![title](https://leanote.com/api/file/getImage?fileId=637e395dab64411710a663c6) * ![title](https://leanote.com/api/file/getImage?fileId=637e3982ab64411710a663c8) ### Implicit Field Supervision For Robust Non-Rigid Shape Matching * 寻找两个非刚体之间的运动映射,用了 Auto-encoder。但传统图形学上应该有不少这种类似的东西。特点是鲁棒性很好。 * ![title](https://leanote.com/api/file/getImage?fileId=637e4019ab64411710a663fd) ### Generalizable Patch-Based Neural Rendering * 这和CVPR那篇ORAL好像啊,好像是同一个组继续水,改进了一点点,主要是增强了泛化能力 * ![title](https://leanote.com/api/file/getImage?fileId=637f15a9ab64411710a66a45) ### ViewFormer: NeRF-free Neural Rendering from Few Images Using Transformers * 尝试在单次 forward 中完成预测,其实就是MVS任务。 * ![title](https://leanote.com/api/file/getImage?fileId=637f18ddab64411710a66a60) ### Transformers as Meta-Learners for Implicit Neural Representations * 这篇尝试用 transformer 学一个初始化(感觉在可控生成这边似乎应该可以试试)。 * ![title](https://leanote.com/api/file/getImage?fileId=637f1aebab64411709a8174a) * ![title](https://leanote.com/api/file/getImage?fileId=637f1afbab64411709a8174b) * ![title](https://leanote.com/api/file/getImage?fileId=637f1b15ab64411710a66a76) ### Sem2NeRF * 为什么我总觉得,这应该早就有人做了呢。 * ![title](https://leanote.com/api/file/getImage?fileId=637f20f5ab64411709a81780) * ![title](https://leanote.com/api/file/getImage?fileId=637f20e0ab64411710a66aa0) ### CompNVS: Novel View Synthesis with Scene Completion * RGB-D 图像映射出来的结果经常有很多洞,然后本文主要就是在补洞。用了一个 GAN-based 的方法。 * ![title](https://leanote.com/api/file/getImage?fileId=637f215cab64411709a81785) * ![title](https://leanote.com/api/file/getImage?fileId=637f219eab64411710a66aa5) ### SinNeRF: Training Neural Radiance Fields on Complex Scenes from a Single Image * 这篇早就看过了 * ![title](https://leanote.com/api/file/getImage?fileId=637f21dcab64411710a66aa6) ### Generative Multiplane Images: Making a 2D GAN 3D-Aware * 还挺有意思,研究如何利用 MPI 把通用 2D GAN改成3D。 * ![title](https://leanote.com/api/file/getImage?fileId=637f343fab64411710a66b39) * ![title](https://leanote.com/api/file/getImage?fileId=637f3474ab64411709a8181b) * **这个是不是也可以搬到 diffusion**? ### ARAH: Animatable Volume Rendering of Articulated Human SDFs * 还是做人体重建的,人体先验+SDF+few shot * ![title](https://leanote.com/api/file/getImage?fileId=637f35b7ab64411710a66b44) * ![title](https://leanote.com/api/file/getImage?fileId=637f35ddab64411710a66b46) ### PREF: Predictability Regularized Neural Motion Fields * 运动预测也做成一个 fields * ![title](https://leanote.com/api/file/getImage?fileId=637f3637ab64411710a66b4a) * ![title](https://leanote.com/api/file/getImage?fileId=637f364fab64411709a8182f) ### NeuMan: Neural Human Radiance Field from a Single Video * 和 CVPR 那篇挺像的,都是从单个视频还原人体+常见 * ![title](https://leanote.com/api/file/getImage?fileId=637f3738ab64411710a66b56) ### AutoAvatar: Autoregressive Neural Fields for Dynamic Avatar Modeling * 建模动态人体 * ![title](https://leanote.com/api/file/getImage?fileId=637f3b71ab64411709a81867) * ![title](https://leanote.com/api/file/getImage?fileId=637f3bb5ab64411710a66b80) ### Pose-NDF: Modeling Human Pose Manifolds with Neural Distance Fields * 新的人体姿态建模方式,可用于体态插值、姿态估计、动作去噪。 * ![title](https://leanote.com/api/file/getImage?fileId=637f397aab64411710a66b6f) * ![title](https://leanote.com/api/file/getImage?fileId=637f39acab64411709a81856) ### DANBO: Disentangled Articulated Neural Body Representations via Graph Neural Networks * 结合骨架学习人体姿态和表面。 * ![title](https://leanote.com/api/file/getImage?fileId=637f3c1aab64411709a8186d) * ![title](https://leanote.com/api/file/getImage?fileId=637f3c22ab64411710a66b88) ### Deforming Radiance Fields with Cages * 太熟了 ### Object-Compositional Neural Implicit Surfaces * 也很熟 ### Neural Correspondence Field for Object Pose Estimation * 感觉看过,给定三维模型先验,然后match 场景中的物体。 * ![title](https://leanote.com/api/file/getImage?fileId=637f522dab64411709a81917) * ![title](https://leanote.com/api/file/getImage?fileId=637f5242ab64411710a66c29) ### RayTran: 3D pose estimation and shape reconstruction of multiple objects from videos with ray-traced transformers * 从三维视频中做形状提取。 * ![title](https://leanote.com/api/file/getImage?fileId=637f53d6ab64411710a66c34) * ![title](https://leanote.com/api/file/getImage?fileId=637f5410ab64411710a66c36) * ![title](https://leanote.com/api/file/getImage?fileId=637f5430ab64411709a81929) ### Gaussian Activated Neural Radiance Fields for High Fidelity Reconstruction & Pose Estimation * 图片相机位置估计+重建 * ![title](https://leanote.com/api/file/getImage?fileId=637f5890ab64411710a66c59) * ![title](https://leanote.com/api/file/getImage?fileId=637f5837ab64411710a66c55) ### NeRF-OSR * Relightning的,讲过了 ### PANDORA: Polarization-Aided Neural Decomposition Of Radiance * 在nerf材质分解中,引入了偏震 * ![title](https://leanote.com/api/file/getImage?fileId=637f59ddab64411709a81951) * ![title](https://leanote.com/api/file/getImage?fileId=637f5ab3ab64411710a66c6c) * ![title](https://leanote.com/api/file/getImage?fileId=637f5acaab64411710a66c6e) * ![title](https://leanote.com/api/file/getImage?fileId=637f5ae7ab64411709a81959) * ![title](https://leanote.com/api/file/getImage?fileId=637f5af7ab64411710a66c6f) ### Relighting4D: Neural Relightable Human from Videos * 从视频重建的人体的Relightning * ![title](https://leanote.com/api/file/getImage?fileId=637f617fab64411710a66ca2) * ![title](https://leanote.com/api/file/getImage?fileId=637f6247ab64411710a66ca8) ### PS-NeRF: Neural Inverse Rendering for Multi-view Photometric Stereo * ![title](https://leanote.com/api/file/getImage?fileId=637f7c2eab64411709a81a4e) * ![title](https://leanote.com/api/file/getImage?fileId=637f7cc2ab64411709a81a53) * ![title](https://leanote.com/api/file/getImage?fileId=637f801dab64411710a66d75) * 这PSNR 有点离谱啊 ### NeILF: Neural Incident Light Field for Physically-based Material Estimation * 也是复杂光照+反射 * ![title](https://leanote.com/api/file/getImage?fileId=637f8150ab64411709a81a73) * ![title](https://leanote.com/api/file/getImage?fileId=637f813aab64411709a81a72) ### RC-MVSNet: Unsupervised Multi-View Stereo with Neural Rendering * MVS任务,添加了 Neural Rendering,提升不是很大 * ![title](https://leanote.com/api/file/getImage?fileId=637f883eab64411710a66daf) * ![title](https://leanote.com/api/file/getImage?fileId=637f8876ab64411709a81aa7) * ![title](https://leanote.com/api/file/getImage?fileId=637f887cab64411709a81aa8) ### ARF: Artistic Radiance Fields * 辐射场的风格化,还挺好看的 * ![title](https://leanote.com/api/file/getImage?fileId=63802864ab64411709a81f04) * 基于特征匹配的做法,我之前还以为是基于最小化风格loss * ![title](https://leanote.com/api/file/getImage?fileId=638028b6ab64411710a67261) * 一次性渲染一张图显存放不下,所以加 Cache 分 Patch * ![title](https://leanote.com/api/file/getImage?fileId=63802983ab64411709a81f0f) ### Unified Implicit Neural Stylization * ![title](https://leanote.com/api/file/getImage?fileId=63803080ab64411710a672c6) * ![title](https://leanote.com/api/file/getImage?fileId=638030b8ab64411710a672c8) * ![title](https://leanote.com/api/file/getImage?fileId=638030d0ab64411709a81f6d) * ![title](https://leanote.com/api/file/getImage?fileId=638030f4ab64411709a81f70) * 另一种方法的风格化,由于预测SDF,所以三维一致性比上面这篇好。大家可以看看它的主页https://zhiwenfan.github.io/INS/。 * ![title](https://leanote.com/api/file/getImage?fileId=638031f6ab64411710a672d3) * 这张图很能说明,如果只用Perceptual Loss的话,一些空气会被一起修改掉。SDF能给出一个明确的修改边界。 ### Implicit Neural Representations for Image Compression * 用 Neural Representations 做图像压缩 * ![title](https://leanote.com/api/file/getImage?fileId=63802b3cab64411709a81f1b) * ![title](https://leanote.com/api/file/getImage?fileId=63802b65ab64411709a81f1f) * 但并不比之前的压缩算法优 * ![title](https://leanote.com/api/file/getImage?fileId=63802ba5ab64411709a81f2b) ### Neural Strands: Learning Hair Geometry and Appearance from Multi-View Images * 做头发重建和渲染的 * ![title](https://leanote.com/api/file/getImage?fileId=63802cdeab64411710a672a3) * 不知道为什么,总觉得这个编辑结果看起来有一点点假(可能因为黑色背景),但至少清晰度是给够了的。 * ![title](https://leanote.com/api/file/getImage?fileId=63802d2fab64411710a672a5) * ![title](https://leanote.com/api/file/getImage?fileId=63802d89ab64411709a81f52) * ![title](https://leanote.com/api/file/getImage?fileId=63802da9ab64411710a672aa) ### DeepShadow: Neural Shape from Shadow * 奇怪的任务,从多个阴影图还原形状 * ![title](https://leanote.com/api/file/getImage?fileId=63802e0dab64411709a81f55) * ![title](https://leanote.com/api/file/getImage?fileId=63802e34ab64411710a672b0) * 或许可以用于地形图还原。 ### LaTeRF: Label and Text Driven Object Radiance Fields * 除了 CLIP 以外再加入标注 * ![title](https://leanote.com/api/file/getImage?fileId=63802e85ab64411709a81f5a) ### Towards Learning Neural Representations from Shadows * 又是一个学阴影的 * ![title](https://leanote.com/api/file/getImage?fileId=63802ee6ab64411710a672b9) ### Minimal Neural Atlas: Parameterizing Complex Surfaces with Minimal Charts and Distortion * 好像是用几个面卷成衣服,输入是点云,感觉可以用于服装生成。 * ![title](https://leanote.com/api/file/getImage?fileId=63803545ab64411710a672ed) * ![title](https://leanote.com/api/file/getImage?fileId=63803575ab64411710a672f0) * ![title](https://leanote.com/api/file/getImage?fileId=63803596ab64411710a672f1)
上一篇:
NeurIPS 2022 NeRF 相关
下一篇:
CLIP-NeRF 后续工作
0
赞
198 人读过
新浪微博
微信
腾讯微博
QQ空间
人人网
提交评论
立即登录
, 发表评论.
没有帐号?
立即注册
0
条评论
More...
文档导航
没有帐号? 立即注册