Skip to content

Latest commit

 

History

History
229 lines (184 loc) · 24.2 KB

visualization.md

File metadata and controls

229 lines (184 loc) · 24.2 KB
Raw

Visualization

We provide a tool for interactive 3D visualization of Data and NAG objects. This allows you to inspect what your Data attributes look like, check whether your model's predictions make sense, and even share your interactive visualizations as simple, portable, HTML files.

Note: If you are not familiar with our Data and NAG structures yet, check the Data documentation.

Let the show() begin

Our show() function rules all visualizations. It can be used as follows:

from src.visualization import show

# Visualize a Data object
show(data)

It is also possible to directly call show() as a Data or NAG method:

# Visualize a Data object
data.show()

# Visualize a NAG object
nag.show()

Datasets inheriting from our BaseDataset class also have a show_examples() method that returns some spherical samples for a given class, randomly picked across the dataset.

# Visualize `max_examples` random spherical samples of size `radius`, 
# centered on points annotated as `label`
dataset.show_examples(label, radius=4, max_examples=5)

Visualizing data attributes

Given a Data or a NAG object, show() will automatically look for and plot some specific point attributes. It generates an interactive Plotly graph with intuitive navigation controls and buttons for accessing different visualization modes 🔘.

The pos attribute is used to plot point positions. The 🔘Position RGB mode will display points colored with respect to their location in the scene.

If the points carry colors in the rgb attribute, the 🔘RGB mode will display them.

If the points carry semantic segmentation labels in the y attribute, the 🔘Labels mode will display them. In addition, show(class_colors=..., class_names=...) can be passed to adjust the color palette and the name displayed when mouse hovering the graph. Similarly, if the points carry panoptic segmentation labels in the obj attribute, the 🔘Panoptic mode will show them. Passing show(stuff_classes=..., num_classes=...) allows for adjusting how thing, stuff, and void classes are displayed.

If the points carry some features in the x attribute, the 🔘Features 3D mode will display them.

Note: This only supports point-wise features stored as 1D or 2D tensors. If the tensor contains only 1 channel, the attribute will be represented with a grayscale colormap. If the tensor contains 2 or 3 channels, these will be represented as RGB, with an additional all-1 channel if need be. If the tensor contains more than 3 channels, a PCA projection to RGB will be shown. In any case, the attribute values will be rescaled with respect to their statistics before visualization, meaning that colors may not compare between two different plots.

If the points carry superpoint partition indices in the super_index attribute, the 🔘Level i mode will display partition level i. Several options can be passed to show() to adjust how partitions are shown: with centroids, with horizontal graph edges, with vertical graph edges, with feature-colored edges, etc. See the exhaustive list of parameters of show() for more details.

If the points carry semantic segmentation predictions in the semantic_pred attribute, the 🔘Semantic Pred. mode will display them. Besides, if both semantic_pred and y can be found in the input, the 🔘Error button can be used to highlight the erroneous predictions, as shown in red here below. Similarly, panoptic_pred attributes can be visualized in the 🔘Panoptic Pred. mode.

Tip 💡: Want to visualize other point attributes stored in your Data object ? Pass them to show(keys=...) to have them displayed ! The same coloring mechanisms as mentioned above for the x attribute will be used.

Points selection for visualization

Our Plotly-based visualization is mostly meant for quick interactive experiments with up to ~10⁵ points. While it is possible to visualize more points at once, Plotly will take a while to generate the plot. We provide several tools for reducing the number of point displayed, hence accelerating the visualization speed.

Say you have the following large-scale visualization.

You can choose to alleviate the visualization time by sampling some points. This can be done in one of two ways:

  • random sampling with respect to a maximum point budget with:
show(max_points=...)
  • grid sampling (ie voxelization) with:
show(voxel=...)

This can make your visualization run faster, at the expense of resolution.

Another strategy is to simply visualize a smaller area. To this end, one can visualize only a spherical crop of the scene with:

show(radius=..., center=...)

We also provide a tool for displaying point selections with:

show(select=..., alpha=...)

This internally calls NAG.select() and Data.select() and applies an alpha opacity to the non-selected points. This is typically useful for visualizing binary masks across the scene. Here is, for example, the mask corresponding to the previous spherical cropping operation.

Sharing your visualization as an HTML

You can export your Plotly figure as an HTML file which can be conveniently opened by virtually anyone with a browser. To this end, simply specify a path to which your HTML should be saved, along with a figure title, if need be:

show(path=..., title=...)

Exhaustive parameter list for show()

Here is the list of supported parameters for show():

Parameter Type Description
input Data or NAG Data or NAG object to visualize
keys List(str) or str By default, the following attributes will be parsed in input for visualization {pos, rgb, y, obj, semantic_pred, obj_pred}. Yet, if input contains other attributes that you want to visualize, these can be passed as keys. This only supports point-wise attributes stored as 1D or 2D tensors. If the tensor contains only 1 channel, the attribute will be represented with a grayscale colormap. If the tensor
figsize int Figure dimensions will be (figsize, figsize/2) if width and height are not specified
width int Figure width
height int Figure height
class_names List(str) Names for point labels found in attributes y and semantic_pred
class_colors List(List(int, int, int)) Colors palette for point labels found in attributes y and semantic_pred
stuff_classes List(int) Semantic labels of the classes considered as stuff for instance and panoptic segmentation. If y and obj are found in the point attributes, the stuff annotations will appear accordingly. Otherwise, stuff instance labeling will appear as any other object
num_classes int Number of valid classes. By convention, we assume y ∈ [0, num_classes-1] are VALID LABELS, while y < 0 AND y >= num_classes ARE VOID LABELS
hide_void_pred bool Whether predictions on points labeled as VOID be visualized
voxel float Voxel size to subsample the point cloud to facilitate visualization
max_points int Maximum number of points displayed to facilitate visualization
point_size int or float Size of point markers
centroid_size int or float Size of superpoint markers
error_color List(int, int, int) Color used to identify mis-predicted points
centroids bool Whether superpoint centroids should be displayed
h_edge bool Whether horizontal edges should be displayed (only if centroids=True)
h_edge_attr bool Whether the edges should be colored by their features found in edge_attr (only if h_edge=True)
h_edge_width float Width of the horizontal edges, if h_edge=True. Defaults to None, in which case point_size will be used for the edge width
v_edge bool Whether vertical edges should be displayed (only if centroids=True and gap is not None)
v_edge_width float Width of the vertical edges, if v_edge=True. Defaults to None, in which case point_size will be used for the edge width
gap List(float, float, float) If None, the hierarchical graphs will be overlaid on the points. If not None, a 3D tensor indicating the offset by which the hierarchical graphs should be plotted
radius float If not None, only visualize a spherical sampling of the input data, centered on center and with size radius. This option is not compatible with select
center List(float, float, float) If radius is provided, only visualize a spherical sampling of the input data, centered on center and with size radius. If None, the center of the scene will be used
select Tuple(int, Tensor) If not None, will call Data.select(*select) or NAG.select(*select) on the input data (depending on its nature) and the coloring schemes will illustrate it. This option is not compatible with radius
alpha float Rules the whitening of selected points, nodes and edges (only if select is not None)
alpha_super float Rules the whitening of superpoints (only if select is not None). If None, alpha will be used as fallback
alpha_stuff float Rules the whitening of stuff points (only if the input points have obj and semantic_pred attributes, and stuff_classes or num_classes is specified). If None, alpha will be used as fallback
point_symbol str Marker symbol used for points. Must be one of {'circle', 'circle-open', 'square', 'square-open', 'diamond', 'diamond-open', 'cross', 'x'}. Defaults to 'circle'
centroid_symbol str Marker symbol used for centroids. Must be one of {'circle', 'circle-open', 'square', 'square-open', 'diamond', 'diamond-open', 'cross', 'x'}. Defaults to 'circle'
colorscale str Plotly colorscale used for coloring 1D continuous features. See https://plotly.com/python/builtin-colorscales for options
path str If not None, path will be used for storing the Plotly figure as an HTML file
title str Title for the Plotly figure
pt_path str Path to save the visualization-ready Data object as a *.pt. In this Data object, the pos and all *color* attributes will be saved, the rest is discarded. This is typically useful for exporting the visualization layers to another visualization tool
kwargs

🚀  Getting started with visualization

See the notebooks/demo_nag.ipynb notebook to play with our visualization tool on a provided NAG, without needing a whole dataset.

For more information, have a look at the code in src.visualization, it is fairly commented and should help you gain a deeper understanding of the visualization mechanisms.