Megascans Food

现代3d感知图像合成基准有两个问题: 1) 它们包含非常
类似的全局几何形状 (例如，人或猫的脸，汽车和椅子)，以及2) 它们的相机较差
覆盖范围。此外，当对象特征 (例如，FFHQ) 时，其中一些 (例如，FFHQ) 包含3d偏差 (例如，
微笑概率、凝视方向、姿势或发型) 与相机位置相关 [6]。作为
结果，这不允许评估模型表示基础几何的能力，并且
使人们更难理解性能是来自方法上的变化还是更好
数据预处理。
为了缓解这些问题，我们引入了两个新的数据集: Megascans植物 (M-Plants) 和Megascans
食物 (M-食物)。为了构建它们，我们从Quixel Megascans6获得 ≈ 1，500模型
来自植物，
蘑菇和食品类别。Megascans是对真实物体的高质量扫描
几乎与真实没有区别。对于蘑菇和植物，我们将它们合并到相同的食物中
类别，因为他们自己的模型太少了。
我们用相机在搅拌机中渲染所有模型，随机均匀分布在
半径3.5的球体和 π/4的视场。在渲染时，我们将每个模型缩放为 [− 1,1] 3
立方体并丢弃那些尺寸小于2的模型。我们呈现128视图
从固定距离到对象中心的每个对象，从整个对象上的均匀采样点
球体 (甚至从下面)。对于M-Plants，我们另外删除了那些少于
平均0.03像素强度 (计算为像素和视图上的平均 α 值)。这是
需要去除小草或树叶，这将占用太少的像素。作为一个
结果，该程序产生了植物类别的1,108模型和食品的199模型
类。

Modern 3D-aware image synthesis benchmarks suffer from two critical issues: 1) they contain highly similar global geometries (e.g., human or cat faces, cars and chairs), and 2) their camera coverage is insufficient. Moreover, some datasets (e.g., FFHQ) exhibit 3D biases such as smile probability, gaze direction, pose or hairstyle that correlate with camera position [6]. As a result, these benchmarks fail to enable rigorous evaluation of a model’s ability to represent underlying geometry, and make it more challenging to distinguish whether performance derives from methodological advancements or superior data preprocessing.

To address these limitations, we propose two novel datasets: Megascans Plants (M-Plants) and Megascans Food (M-Food). To build them, we acquired approximately 1,500 models from Quixel Megascans6 across plant, mushroom and food categories. Megascans features high-quality scans of real-world objects that are nearly indistinguishable from their physical counterparts. Due to the insufficient number of standalone models for mushrooms and plants, we grouped these two categories into the same food category.

We rendered all models in Blender, with cameras randomly and uniformly distributed on a sphere of radius 3.5 and a field of view of π/4. During rendering, we scaled each model to fit within the [-1, 1]^3 cube and discarded models with a size smaller than 2. We rendered 128 views for each object at a fixed distance to the object’s center, with sampling points uniformly distributed across the entire sphere (including from below). For M-Plants, we further removed models with an average pixel intensity lower than 0.03, calculated as the average alpha value across all pixels and views. This step is necessary to eliminate tiny grass or leaves that occupy too few pixels. Ultimately, this pipeline yields 1,108 models for the plant category and 199 models for the food category.

搜集汇总

背景与挑战

背景概述

Megascans Food数据集是为了解决现有3D感知图像合成基准在全局几何相似性和相机覆盖方面的局限性而创建的。它基于Quixel Megascans的高质量扫描模型，包含199个食品类别模型，并通过Blender渲染生成每个对象128个均匀分布的视图。