3D Preview Thumbnail Pipeline

The 3D Preview Thumbnail pipeline generates animated GIF or static image previews from 3D files, providing visual thumbnails for assets in the VAMS web interface. It supports a wide range of mesh, point cloud, CAD, and USD file formats. The pipeline uses CPU-based headless rendering with PyVista, VTK, and Xvfb inside an AWS Batch Fargate container.

Supported Formats

Mesh Formats

Format	Extension	Library
PLY	.ply	Trimesh
STL	.stl	Trimesh
OBJ	.obj	Trimesh
GLB	.glb	Trimesh
GLTF	.gltf	Trimesh (with external dependency download)
FBX	.fbx	Trimesh
DRC (Draco)	.drc	Trimesh

Point Cloud Formats

Format	Extension	Library
LAS	.las	laspy
LAZ	.laz	laspy + laszip
E57	.e57	pye57
PTX	.ptx	Open3D
PCD	.pcd	Open3D
FLS	.fls	Open3D
FWS	.fws	Open3D

CAD Formats

Format	Extension	Library
STEP	.stp, .step	CadQuery / Open CASCADE

USD Formats

Format	Extension	Library
USD	.usd	OpenUSD (pxr)
USDA	.usda	OpenUSD (pxr)
USDC	.usdc	OpenUSD (pxr)
USDZ	.usdz	OpenUSD (pxr)

Architecture

Processing Steps

1. Download -- The container downloads the input file from the asset Amazon S3 bucket. For GLTF files, external dependencies (buffers, textures) are also downloaded.
2. Load -- A format-specific handler loads the file into a PyVista PolyData object. The handler is selected based on file extension: mesh, point cloud, CAD, or USD.
3. Normalize -- The pipeline detects and normalizes the coordinate up-axis to Y-up for consistent rendering. Z-up formats (LAS, LAZ, E57, STL, etc.) are rotated automatically. Variable formats use a bounding-box heuristic.
4. Render -- PyVista generates rotating preview frames using headless off-screen rendering via Xvfb. If animated rendering fails, the pipeline falls back to a single static frame.
5. Save -- Multiple frames are assembled into an animated GIF with automatic size optimization. A single frame is saved as a JPEG. The pipeline ensures the output stays under a size limit for efficient web display.
6. Upload -- The preview file is uploaded to the asset bucket alongside the source file, preserving the relative directory structure.

Output Files

The pipeline generates preview files with the following naming convention:

<original_filename>.previewFile.gif    (animated, multi-frame)
<original_filename>.previewFile.jpg    (static, single-frame fallback)

These files are written to outputS3AssetFilesPath, which maps to the asset bucket. The relative subdirectory from the input path is preserved so the VAMS process-output step can locate and register the previews correctly.

Relative Path Preservation

If the input file is at <assetId>/subfolder/model.glb, the preview is written to <assetId>/subfolder/model.glb.previewFile.gif. The assetId is threaded through the pipeline from the workflow state to ensure correct path computation.

Example Output

The following are example animated GIF previews generated by the pipeline across different file types:

GLTF Mesh	USD Scene	Point Cloud (E57)	Point Cloud (LAZ)
Avocado.gltf	gramophone.usdz	pump.e57	autzen.laz

Configuration

Enable this pipeline in infra/config/config.json:

{
    "app": {
        "pipelines": {
            "usePreview3dThumbnail": {
                "enabled": true,
                "autoRegisterWithVAMS": true,
                "autoRegisterAutoTriggerOnFileUpload": true
            }
        }
    }
}

Configuration Options

Option	Default	Description
enabled	false	Deploy the 3D thumbnail pipeline infrastructure. Enables the global VPC.
autoRegisterWithVAMS	false	Automatically register the pipeline and workflow during CDK deployment.
autoRegisterAutoTriggerOnFileUpload	false	Automatically trigger the pipeline when supported 3D files are uploaded.

License Notice

This pipeline is disabled by default because it depends on libraries with LGPL licenses (CadQuery/Open CASCADE for STEP file support). Review the requirements.txt file in the container directory and consult your legal team before enabling this pipeline. Other format handlers use MIT-licensed or Apache-licensed libraries.

Input Parameters

When executing the pipeline manually or via API, you can pass the following input parameters:

Parameter	Type	Default	Description
overwriteExistingPreviewFiles	boolean	true	When true, regenerates preview files even if they already exist for the input file. When false, the pipeline skips files that already have a preview.

Example Input Parameters

{
    "inputParameters": "{\"overwriteExistingPreviewFiles\": true}"
}

Limits and Constraints

Constraint	Value
Maximum input file size	100 GB
Container ephemeral storage	200 GiB
Point cloud downsampling threshold	20 million points
Rendering method	CPU-based (no GPU required)

Large Point Clouds

Point clouds exceeding 20 million points are automatically downsampled for rendering performance. The original file is not modified; only the rendering input is reduced.

Prerequisites

VPC

This pipeline runs on AWS Batch with AWS Fargate. Enabling it automatically enables the global VPC and creates the necessary VPC endpoints for AWS Batch, Amazon ECR, and Amazon ECR Docker.

Container Image

The container image is built during CDK deployment from backendPipelines/preview/3dThumbnail/container/Dockerfile. It is based on Python 3.12 slim and includes:

• PyVista / VTK -- 3D rendering engine (MIT license)
• Trimesh -- Mesh loading for PLY, STL, OBJ, GLB, GLTF, FBX, DRC (MIT license)
• laspy -- LAS/LAZ point cloud reading (BSD license)
• pye57 -- E57 point cloud reading (MIT license)
• Open3D -- Additional point cloud formats: PTX, PCD, FLS, FWS (MIT license)
• CadQuery -- STEP/STP CAD file support (LGPL license)
• OpenUSD (pxr) -- USD format support (Modified Apache 2.0 license)
• Xvfb -- Virtual framebuffer for headless rendering

How It Works

1. The workflow triggers the vamsExecute Lambda function with the input file path, all S3 output paths, and the assetId.
2. The constructPipeline Lambda function builds a PREVIEW_3D_THUMBNAIL stage definition, directing output to outputS3AssetFilesPath (the asset bucket).
3. AWS Batch submits an AWS Fargate job. The container starts Xvfb for headless rendering, then runs the preview pipeline.
4. The container downloads the input file, loads it with the appropriate format handler, normalizes the up-axis, renders rotating frames, saves the output as GIF or JPEG, and uploads it to Amazon S3.
5. AWS Step Functions receives the task token callback, and the process-output step registers the preview file in VAMS.

Related Resources

• Pipeline System Overview
• Potree Point Cloud Viewer Pipeline -- interactive point cloud visualization (complementary to thumbnail previews)
• CAD/Mesh Metadata Extraction Pipeline -- extracts metadata from similar file formats