官网地址
https://drive.google.com/drive/folders/128yBriW1IG_3NJ5Rp7APSTZsJqdJdfc1
文件夹内容
nerf_synthetic README.txt
This directory contains a dataset of synthetically rendered images that were used in "NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis". Stats: + 8 Scenes + 100 Training images + 100 Validation images + 200 Test images + Images are 800x800 Structure: SCENE_NAME -train r_*.png -val r_*.png -test r_*.png r_*_depth_0000.png r_*_normal_0000.png transforms_train.json transforms_val.json transforms_test.json Transform json details: camera_angle_x: The FOV in x dimension frames: List of dictionaries that contain the camera transform matrices for each image. Attribution: The renders are from modified blender models located on blendswap.com chair by 1DInc (CC-0): https://www.blendswap.com/blend/8261 drums by bryanajones (CC-BY): https://www.blendswap.com/blend/13383 ficus by Herberhold (CC-0): https://www.blendswap.com/blend/23125 hotdog by erickfree (CC-0): https://www.blendswap.com/blend/23962 lego by Heinzelnisse (CC-BY-NC): https://www.blendswap.com/blend/11490 materials by elbrujodelatribu (CC-0): https://www.blendswap.com/blend/10120 mic by up3d.de (CC-0): https://www.blendswap.com/blend/23295 ship by gregzaal (CC-BY-SA): https://www.blendswap.com/blend/8167
Realistic Synthetic 360°
如图 2( a) 所示,Realistic Synthetic 360°是为评 估 NeRF 方法有效性提供的合成 数据集,包 括 Chair,Drums,Ficus,Hotdog,Lego,Materials,Mic 和 Ship 等 8 个由 Blender 软件精细建模的复杂非朗伯 场景. 每个场景的图片由 Blender 的环形路径追踪 插件采集,其中的 6 个场景是从上半球方向环绕拍 摄的,另外 2 个场景是从整个球形的其他方向采集. 选取每个场景的 100 张图片用神经网络算法进行训 练,其中的 13 张图片用于测试. 该数据集所有图片 的分辨率为 800 × 800.
Real Forward-Facing
如图 2 所示,Real Forward-Facing 数据集包括 Room[6],Fern[4],Leaves[6],Fortress[4],Orchids[4], Flower[4],Trex[4] 和 Horns[6] 等 8 个真实世界的场 景. 与 Realistic Synthetic 360°数据集不同,该数据集 的图片是使用智能手机面对着场景上下左右平移拍 摄的,即采用 Forward-Facing 视角拍摄. 由于是手持 拍摄的真实场景,每张图片对应的摄像机参数由 colmap[12 - 13]估 计 得 到. 每个场景的图片数量为 20 ~ 62 不等,其中 1 /8 的图片可作为测试集,其他 图片作为训练集. 该数据集所有图片的分辨率为 1 008 ×756.