A Scene understanding, Accessibility, Navigation, Pathfinding, & Impediment avoidance dataset – Google Analysis Weblog

As most individuals navigate their on a regular basis world, they course of visible enter from the atmosphere utilizing an eye-level perspective. In contrast to robots and self-driving vehicles, individuals have no “out-of-body” sensors to assist information them. As an alternative, an individual’s sensory enter is totally “selfish”, or “from the self.” This additionally applies to new applied sciences that perceive the world round us from a human-like perspective, e.g., robots navigating by way of unknown buildings, AR glasses that spotlight objects, or assistive technology to help people run independently.

In pc imaginative and prescient, scene understanding is the subfield that research how seen objects relate to the scene’s 3D construction and format by specializing in the spatial, useful, and semantic relationships between objects and their atmosphere. For instance, autonomous drivers should perceive the 3D construction of the highway, sidewalks, and surrounding buildings whereas figuring out and recognizing road indicators and cease lights, a job made simpler with 3D information from a particular laser scanner mounted on the highest of the automobile slightly than 2D photographs from the motive force’s perspective. Robots navigating a park should perceive the place the trail is and what obstacles may intrude, which is simplified with a map of their environment and GPS positioning information. Lastly, AR glasses that assist customers discover their approach want to grasp the place the person is and what they’re .

The pc imaginative and prescient group usually research scene understanding duties in contexts like self-driving, the place many different sensors (GPS, wheel positioning, maps, and so on.) past selfish imagery can be found. But most datasets on this house don’t focus completely on selfish information, so they’re much less relevant to human-centered navigation duties. Whereas there are many self-driving targeted scene understanding datasets, they’ve restricted generalization to selfish human scene understanding. A complete human selfish dataset would assist construct methods for associated functions and function a difficult benchmark for the scene understanding group.

To that finish, we current the Scene understanding, Accessibility, Navigation, Pathfinding, Obstacle avoidance dataset, or SANPO (additionally the Japanese phrase for ”brisk stroll”), a multi-attribute video dataset for outside human selfish scene understanding. The dataset consists of actual world information and artificial information, which we name SANPO-Actual and SANPO-Artificial, respectively. It helps all kinds of dense prediction duties, is difficult for present fashions, and consists of actual and artificial information with depth maps and video panoptic masks during which every pixel is assigned a semantic class label (and for some semantic lessons, every pixel can be assigned a semantic occasion ID that uniquely identifies that object within the scene). The actual dataset covers numerous environments and has movies from two stereo cameras to help multi-view strategies, together with 11.4 hours captured at 15 frames per second (FPS) with dense annotations. Researchers can obtain and use SANPO here.

3D scene of an actual session constructed utilizing the offered annotations (segmentation, depth and digital camera positions). The highest heart video reveals the depth map, and the highest proper reveals the RGB or semantic annotations.


SANPO-Actual is a multiview video dataset containing 701 periods recorded with two stereo cameras: a head-mounted ZED Mini and a chest-mounted ZED-2i. That’s 4 RGB streams per session at 15 FPS. 597 periods are recorded at a decision of 2208×1242 pixels, and the rest are recorded at a decision of 1920×1080 pixels. Every session is roughly 30 seconds lengthy, and the recorded movies are rectified utilizing Zed software and saved in a lossless format. Every session has high-level attribute annotations, digital camera pose trajectories, dense depth maps from CREStereo, and sparse depth maps offered by the Zed SDK. A subset of periods have temporally constant panoptic segmentation annotations of every occasion.

The SANPO information assortment system for gathering real-world information. Proper: (i) a backpack with ZED 2i and ZED Mini cameras for information assortment (backside), (ii) the within of the backpack exhibiting the ZED field and battery pack mounted on a 3D printed container (center), and (iii) an Android app exhibiting the reside feed from the ZED cameras (prime). Left: The chest-mounted ZED-2i has a stereo baseline of 12cm with a 2.1mm focal size, and the head-mounted ZED Mini has a baseline of 6.3cm with a 2.1mm focal size.

Temporally constant panoptic segmentation annotation protocol

SANPO consists of thirty totally different class labels, together with numerous surfaces (highway, sidewalk, curb, and so on.), fences (guard rails, partitions,, gates), obstacles (poles, bike racks, timber), and creatures (pedestrians, riders, animals). Gathering high-quality annotations for these lessons is a gigantic problem. To supply temporally constant panoptic segmentation annotation we divide every video into 30-second sub-videos and annotate each fifth body (90 frames per sub-video), utilizing a cascaded annotation protocol. At every stage, we ask annotators to attract borders round 5 mutually unique labels at a time. We ship the identical picture to totally different annotators with as many levels because it takes to gather masks till all labels are assigned, with annotations from earlier subsets frozen and proven to the annotator. We use AOT, a machine studying mannequin that reduces annotation effort by giving annotators computerized masks from which to begin, taken from earlier frames in the course of the annotation course of. AOT additionally infers segmentation annotations for intermediate frames utilizing the manually annotated previous and following frames. Total, this strategy reduces annotation time, improves boundary precision, and ensures temporally constant annotations for as much as 30 seconds.

Temporally constant panoptic segmentation annotations. The segmentation masks’s title signifies whether or not it was manually annotated or AOT propagated.


Actual-world information has imperfect floor fact labels as a result of {hardware}, algorithms, and human errors, whereas artificial information has near-perfect floor fact and might be personalized. We partnered with Parallel Domain, an organization specializing in lifelike artificial information technology, to create SANPO-Artificial, a high-quality artificial dataset to complement SANPO-Actual. Parallel Area is expert at creating handcrafted artificial environments and information for machine studying functions. Because of their work, SANPO-Artificial matches real-world seize situations with digital camera parameters, placement, and surroundings.

3D scene of an artificial session constructed utilizing the offered annotations (segmentation, depth and odometry). The highest heart video reveals the depth map, and the highest proper reveals the RGB or semantic annotations.

SANPO-Artificial is a top quality video dataset, handcrafted to match actual world eventualities. It incorporates 1961 periods recorded utilizing virtualized Zed cameras, evenly break up between chest-mounted and head-mounted positions and calibrations. These movies are monocular, recorded from the left lens solely. These periods range in size and FPS (5, 14.28, and 33.33) for a mixture of temporal decision / size tradeoffs, and are saved in a lossless format. All of the periods have exact digital camera pose trajectories, dense pixel correct depth maps and temporally constant panoptic segmentation masks.

SANPO-Artificial information has pixel-perfect annotations, even for small and distant cases. This helps develop difficult datasets that mimic the complexity of real-world scenes. SANPO-Artificial and SANPO-Actual are additionally drop-in replacements for one another, so researchers can examine area switch duties or use artificial information throughout coaching with few domain-specific assumptions.

An excellent sampling of actual and artificial scenes.


Semantic lessons

We designed our SANPO taxonomy: i) with human selfish navigation in thoughts, ii) with the aim of being fairly simple to annotate, and iii) to be as shut as potential to the present segmentation taxonomies. Although constructed with human selfish navigation in thoughts, it may be simply mapped or prolonged to different human selfish scene understanding functions. Each SANPO-Actual and SANPO-Artificial function all kinds of objects one would count on in selfish impediment detection information, equivalent to roads, buildings, fences, and timber. SANPO-Artificial features a broad distribution of hand-modeled objects, whereas SANPO-Actual options extra “long-tailed” lessons that seem sometimes in photographs, equivalent to gates, bus stops, or animals.

Distribution of photographs throughout the lessons within the SANPO taxonomy.

Occasion masks

SANPO-Artificial and a portion of SANPO-Actual are additionally annotated with panoptic occasion masks, which assign every pixel to a category and occasion ID. As a result of it’s usually human-labeled, SANPO-Actual has a lot of frames with usually lower than 20 cases per body. Equally, SANPO-Artificial’s digital atmosphere presents pixel-accurate segmentation of most original objects within the scene. Because of this artificial photographs often function many extra cases inside every body.

When contemplating per-frame occasion counts, artificial information often options many extra cases per body than the labeled parts of SANPO-Actual.

Comparability to different datasets

We examine SANPO to different essential video datasets on this area, together with SCAND, MuSoHu, Ego4D, VIPSeg, and Waymo Open. A few of these are meant for robotic navigation (SCAND) or autonomous driving (Waymo) duties. Throughout these datasets, solely Waymo Open and SANPO have each panoptic segmentations and depth maps, and solely SANPO has each actual and artificial information.

Comparability to different video datasets. For stereo vs mono video, datasets marked with ★ have stereo video for all scenes and people marked ☆ present stereo video for a subset. For depth maps, ★ signifies dense depth whereas ☆ represents sparse depth, e.g., from a lower-resolution LIDAR scanner.

Conclusion and future work

We current SANPO, a large-scale and difficult video dataset for human selfish scene understanding, which incorporates actual and artificial samples with dense prediction annotations. We hope SANPO will assist researchers construct visible navigation methods for the visually impaired and advance visible scene understanding. Further particulars can be found in the preprint and on the SANPO dataset GitHub repository.


This dataset was the result of laborious work of many people from numerous groups inside Google and our exterior associate, Parallel Area.

Core Group: Mikhail Sirotenko, Dave Hawkey, Sagar Waghmare, Kimberly Wilber, Xuan Yang, Matthew Wilson

Parallel Area: Stuart Park, Alan Doucet, Alex Valence-Lanoue, & Lars Pandikow.

We might additionally wish to thank following group members: Hartwig Adam, Huisheng Wang, Lucian Ionita, Nitesh Bharadwaj, Suqi Liu, Stephanie Debats, Cattalyya Nuengsigkapian, Astuti Sharma, Alina Kuznetsova, Stefano Pellegrini, Yiwen Luo, Lily Pagan, Maxine Deines, Alex Siegman, Maura O’Brien, Rachel Stigler, Bobby Tran, Supinder Tohra, Umesh Vashisht, Sudhindra Kopalle, Reet Bhatia.

Leave a Reply

Your email address will not be published. Required fields are marked *