Project Description
Participants will download the Reality Composer app and take 3D scans of household objects as defined in the guidelines below.
Project Requirements
Device Requirement:
iPhone 12 Pro with LiDAR or later (12 Pro, 13 Pro, 14 Pro,15 Pro, and Pro Max versions are also applicable).
iPad Pro 4th generation with LiDAR or later.
It requires iOS 17 or later.
A selfie sticks for the iPhone to stabilize the video during the movement (just a recommendation, it is not a must).
Object Spec (criteria for selecting the objects):
70% nominal objects (details and examples are provided later here)
Nominal object size is within 7cm (3 inch) to 180cm (6 feet).
Static object which won’t bend or deform during scan
For deformable object like a t-shirt is OK but we should avoid flipping these objects, even if recommended by the app.
Choose objects with varied surface texture.
Choose objects which are non-reflective, with matte surfaces.
Avoid objects that are very thin in one dimension.
30% challenging objects (details and examples are provided later here)
Smaller than 7cm, larger than 180cm or 1 aggressor (part of the object are low textured, reflective, transparent or translucent).
Do not take scans of boxes that items came in (e.g. a box a coffee maker came in); food boxes are subject to quality review.
Environment Spec (for nominal and challenging objects):
Diffuse consistent lighting.
No strong highlights / spotlights.
Please note that the app will often say “more light required” – if you have good lighting already, you may ignore this.
Please do not add spotlights or bright lights to try to resolve it as that may create other quality issues.
No hard shadows.
Enough space to move around the object.
If scanning outdoors, cloudy day is the best.
NOTE: A max of 5 photos per scan can have a quality issue like bright lighting or blurriness (but more than 5 will cause a rejection of the file).
Tool Orientation
The object scanning will be completed through Reality Composer application on the specified Apple devices.
Download the "Reality Composer" from the App Store.
Download "iZip" application from the App store (this will be needed to zip the files before submitting them).
Video Tutorial - How to Scan the Object
Please watch the following demonstration on how to scan the object to complete the below three segments accordingly.
Video Tutorial: You can click here to watch the tutorial.
Important General Guidelines (General Acceptance Criteria)
No PII (Personal Identifiable Information), so the pictures taken during the scan cannot include any person in the background.
Don’t collect any NSFW data (Not Safe for Work).
Capture process should follow the Reality Composer tutorial with environment, lighting and scanning.
Move slowly around the object.
Enough space to move around the object.
Flip the object on the side when it’s suggested by the application.
Diffuse consistent lighting.
Follow the object category distribution.
Follow the object property distribution. (70% nominal, 30% challenging).
No duplicate objects.
No empty or corrupted files.
Please see additional details and examples for nominal objects and challenging objects.
Ensure images are not truncated, and all images fully capture the object’s parts.
Avoid old, broken, and rusted objects from the junkyard
Object Scanning Steps
iPhone/iPad Position:
Use the iPhone/iPad/ camera in a vertical position.
Move closer to the object to make sure it is centered in the camera frame.
The object must take at least 50% of the frame.
You start from a higher position away from the object taking into consideration the 50% rule.
Segment 1
The object is placed in a standard position.
Start to 3D Scan the object from point A (the point where you are starting the recording). The camera needs to cover the sides and upper part of the object.
Start to move with camera to around the object, the movement should be from the right side (the object is still centered in the frame).
The movement needs to be really slow around the object.
Complete the round/circle movement around the object till you reach point A one more time.
Segment 2
Flippable vs. Non-Flippable Objects: The Reality Composer app will recommend whether to flip an object – follow the on screen instructions.
IMPORTANT: If the object is deformable upon moving (e.g. a sweatshirt), please do NOT flip it, even if the app recommends this)
Turn the object on the side (or, if not flipping, take the picture from a lower level).
Repeat the steps in Segment 1 (follow the on screen instructions).
Move closer to the object to make sure it is centered in the camera frame.
Start to 3D scan the object from point A (the point where you are starting the recording).
The camera needs to cover the sides and upper part of the object.
Start to move with camera to around the object, the movement should be from the right side (the object is still centered in the frame).
Complete the round/circle movement around the object till you reach point A one more time.
Segment 3
Turn the object on the side (or, if not flipping, take the picture from a higher level) and make sure that the areas you captured before can still be seen in the position of the object.
Repeat the steps in Segment 1 (follow the on screen instructions).
Move closer to the object to make sure it is centered in the camera frame.
Start to 3D scan the object from point A (the point where you are starting the recording).
The camera needs to cover the sides and upper part of the object.
Start to move with camera to around the object, the movement should be from the right side (the object is still centered in the frame).
Complete the round/circle movement around the object till you reach point A one more time.
How to Save the File
Save the file as “3D Model and Photos”.
Note – do not select “Model Only” option, this will be rejected.
The image count captured should not be lower than 72 or higher than 200.
The app will show the final 3D image, check the quality before submitting.
If image count is less than 68, rescan the object, otherwise it will be rejected in the QA process, and the file will be classified as "Broken Zip file"
Data Collection Specifications (What to Collect)
General Object Categories:
Furniture: chairs, sofas, tables, etc.
Home Decor: lamps, vases, sculptures, pottery, etc.
Electronics: speakers, game consoles, etc.
Toys: action figures, dolls, toy cars, etc.
Tools: drills, hammers, wrenches, screwdriver, etc.
Kitchenware: pots, cups, pans, appliances, etc.
Nature: indoor plants, flowers, trees, fruits, etc.
Musical instruments: guitars, drums, violines.
Sports equipment: rackets, helmets, all, types of balls and bats, skateboard
Miniatures: dollhouse, model cars, figurines, keychains, fridge magnets, miniature Eiffel Tower, etc.
Stationaries: paperweights, books/notebooks.
Movie props: hats, tools, set pieces like walls, floors, doors, windows, etc.
Others.
Click On -> Priority Object List
To see visual examples and more details on the priority objects.
Avoid Collecting (Rejected Objects):
Please avoid scanning any of the below objects, in case any were found in data collection/scanning set will be rejected.
Shoes
Bottles
Cans
Boxes
Books
Tubes (sunblock tubes, toothpaste tubes, face wash tubes, etc.)
Rocks
Cement blocks
Wood
Towels
Remote Controls
Folded clothes
!! Important Note: Upon submission, please ensure you are selecting the correct category for your objects.
Please avoid choosing the "Others" category unless absolutely applicable.
B. Challenging Objects (30%):
Challenging objects have 1 aggressor.
Aggressors of challenging objects with target distributions:
Very small - less than 3 inch
Very big - larger than 6 feet
Textureless
Reflective
Transparent
Translucent
How to Avoid Rejection for Challenging Objects:
Each object should have a successfully reconstructed 3D model.
Sharp texture, and not a blurry model or images.
No missing details like large protruding parts or missing parts.
Bottom can be missing but all other sides are scanned (no flip required, follow capture guidance in RC app).
Sufficient coverage: at least 72 images, all 3 rings of capture from different elevations with evenly distributed capturing angles.
Need to fill the ring on the UI for each elevation.
Small, localized reconstruction quality issues are acceptable.