采样¶
参考
- 面板:
Sampling is the process of tracing rays from the camera into the scene and bouncing them around until they reach a light source such as a Light object, an emissive mesh, or the world background. The algorithm for this is known as the integrator.
- 视图(最大)采样
The number of paths to trace per pixel in the 3D Viewport (when using the Rendered shading mode). Setting this value to zero enables indefinite sampling.
- 渲染(最大)采样
The number of paths to trace per pixel in the final render. A higher number results in a cleaner image at the cost of a longer render time.
- 时间限制
Stops rendering if the time exceeds the limit, even if the desired sample count hasn't been reached yet. Setting this value to zero disables the limit.
Note
这个时间限制不包括渲染前的处理时间,只包括渲染时间。
自适应采样¶
If the Noise Threshold checkbox is enabled, Cycles will use adaptive sampling, cutting short the sampling process in areas that have become less noisy than the specified threshold value -- in other words, not wasting time further refining areas that already look good enough. For example, hair on a character may need many samples, but the background may only need few.
- 噪波阈值
The threshold to decide whether to continue sampling a pixel. Typical values are in the range from 0.1 to 0.001, with lower values meaning better quality but longer render times. Setting this to 0 makes Cycles guess an automatic value based on the total sample count.
- 最小采样
The minimum number of samples a pixel receives before adaptive sampling is applied. When set to 0 (default), Cycles automatically picks a value determined by the Noise Threshold.
降噪¶
Denoising uses a specialized algorithm to get a less noisy image without requiring more samples.
See also
If you enable the Denoising Data render pass, you can alternatively denoise in the compositing step using the 降噪节点.
- 降噪(视图)
Whether to perform denoising for the 3D Viewport in the Rendered shading mode.
- 降噪(渲染)
Whether to perform denoising for the final rendered image.
- 降噪器
The algorithm to use.
- 自动:
若可用,使用 GPU 加速降噪,以求更好的性能。OpenImageDenoise 优先于 OptiX 。
- OpenImageDenoise:
使用英特尔的 AI 降噪器 Open Image Denoise。通常可提供最高质量。
- OptiX:
使用英伟达的 OptiX AI 降噪器。有些英伟达的旧 GPU 支持 GPU 加速,而英特尔的 OpenImageDenoise 不支持。
仅适用于在 Cycles 渲染设备用户偏好设置中配置的 NVIDIA GPU。
- 通道
Controls which Render Passes the denoiser should use as input. Generally, the more passes the denoiser has access to, the better the result. It is recommended to use at least Albedo as None can blur out details, especially at lower sample counts.
- 无:
使用颜色数据对图像进行去噪。
- 反照率:
使用颜色和反照率数据对图像进行去噪。
- 反照率 + 法向:
使用颜色、反照率和法向通道对图像进行降噪。
- 预过滤 OpenImageDenoise
Controls whether prefiltering is applied to the Passes before denoising. Visible only when using OpenImageDenoise.
- 无:
不对输入通道进行任何预过滤。这个选项保留了最多的细节,也是最快的,但是假设输入的通道是无噪点的,这可能需要很高的采样数。如果输入的通道不是无噪点的,那么去噪后的图像中会有噪点。
- 快速:
假设输入通道不是无噪声的,但不会对输入通道应用预过滤。此选项比 精确 更快,但会产生更模糊的结果。
- 精确:
Prefilters the input passes before denoising to reduce noise. This option usually produces more detailed results than Fast, but with increased processing time.
- 品质 OpenImageDenoise
整体降噪质量。这个选项只有选用 OpenImageDenoise 时才可见。
- 高:
Produces the highest quality at the cost of time.
- 均衡:
兼顾性能和品质。
- 快速:
Sacrifices quality for speed (ideal for viewport rendering).
- 起始采样
Sample at which to start denoising in the 3D Viewport.
- 使用 GPU
在 GPU 上执行降噪。这比在 CPU 上快得多,但需要额外的 GPU 内存。当大型场景需要更多 GPU 内存时,可以禁用该选项。
See GPU Rendering for details on supported GPUs.
路径引导¶
Path guiding helps reduce noise in scenes where finding a path to a light source is difficult for regular path tracing, for example when a room is lit through a small door opening. Important light directions are learned over time, improving as more samples are taken.
Note
路径引导只有在使用CPU渲染的时候才可用。
尽管有些场景中,路径引导可以帮助渲染焦散,但它不是转为渲染复杂焦散效果而设计的,因为这种场景会更难学习路径。
See also
门户 to guide the path tracing manually.
- 训练采样
用来学习的最大采样数,设置为 0 则会一直学习直到渲染结束。一般来说,128 到 256 的采样数就足够得到高质量的路径引导了,更高的值只能得到引导精度的小幅提升,但也会延长渲染时间。
- 曲面
对表面的漫反射和光泽部分启用路径引导。
- 体积/音量
在体积内部启用路径引导。
灯光¶
- 灯光树
建立灯光树来加速场景灯光的采样,会考虑光源距离的预估的亮度。此项可以显著地减少噪点,但也会使单次渲染变长少许。
Certain lighting properties are not accounted for in the light tree. This includes custom falloff, ray visibility, and complex shader node setups including textures. This can result in an increase in noise in some scenes that make use of these features.
- 光照阈值
Probabilistically terminates light samples when the light contribution is below the threshold. This avoids wasting time on lights that contribute little to the image. Zero disables the test.
高级¶
- 图案
积分器使用的随机采样模式。
- 自动:
Use Blue-Noise (see below), but with a tweak for viewport rendering to get better quality in the first sample. This is useful for interactive changes (panning the view, moving an object...) where the render is constantly restarting.
- 经典:
使用预先计算的Owen加扰Sobol表进行随机采样。
- 蓝噪点:
Use a blue-noise pattern, which optimizes the frequency distribution. If the full number of samples is rendered, the output typically appears smoother than Classic despite not actually reducing the overall noise.
- 随机种
积分器使用的随机种,可用来得到不同的噪点随机态。
- 使用动态种子 (时钟图标)
为每一帧改变随机种。建议在渲染动画时启用,这样可以在每一帧改变噪点的随机态,让观众不容易注意到噪点。
- 置乱距离
A technique that reduces the randomness between pixels in an attempt to improve GPU rendering performance, at the cost of potential artifacts. Not compatible with the Blue-Noise sampling pattern.
- 自动
Adapts the scrambling distance based on the sample count.
- 视图
Uses the Scrambling Distance adjustment for viewport rendering. This will make rendering faster but may cause flickering.
- 乘数
A multiplier for the scrambling distance. Values below one will reduce the distance, having the potential to improve GPU rendering performance but increase the visibility of artifacts.
- 最小光线反弹
Minimum number of light bounces for each path, after which the integrator uses Russian Roulette to terminate paths that contribute less to the image. Setting this higher gives less noise, but may also increase render time considerably. For a low number of maximum bounces, it is strongly recommended to set this minimum to the same value.
- 最小透明反弹
Minimum number of transparent bounces (more specifically "passthroughs"). Setting this higher reduces noise in the first bounces, but can also be less efficient for more complex geometry like hair and volumes.
- 层采样
If any view layers have Sample Overrides configured, this option specifies how to use them.
- 使用:
Allow view layers to override the scene-level sample count.
- 限界:
Allow the overrides as long as they don't exceed the scene-level sample count.
- 忽略:
Ignore the overrides.
- Sample Subset
Only render a subset of the samples. Multiple subset renders can be combined into a full one by running the following in the Python Console:
bpy.ops.cycles.merge_images(input_filepath1=r"1.exr", input_filepath2=r"2.exr", output_filepath=r"combined.exr")
A typical use case is to distribute the rendering of a single frame over multiple machines. Say you want to render 2048 samples in total, but split this work over two machines because doing it on one would take too long:
On both machines, set the (Max) Samples to 2048, disable Denoise, and enable Sample Subset.
On the first machine, set Offset to 0 and Length to 1024.
On the second machine, set Offset to 1024 and Length to 1024.
Run the 1024-sample render on each machine, then combine the results as described above to get the equivalent of a single 2048-sample render.
- 偏移量
The 0-based index of the first sample in the subset.
- 长度
The number of samples in the subset. While this overrides (Max) Samples in terms of the samples that will get rendered, it's still important to set (Max) Samples to the total number of samples that will be rendered across all subsets -- otherwise, the subsets will have incompatible noise and combining them will give a worse result.