Tile-based caching is a content delivery technique that spatially segments a high-resolution video frame into independent, decodable rectangular regions called tiles. Instead of caching the entire omnidirectional frame, the system pre-fetches and stores only the viewport-dependent tiles corresponding to a user's predicted field of view. This mechanism leverages HTTP Adaptive Streaming protocols, where each tile is encoded at multiple quality levels, allowing the edge cache to serve high-bitrate data for the visible region while discarding or storing low-quality versions of unseen areas, dramatically reducing backhaul consumption.
Glossary
Tile-Based Caching

What is Tile-Based Caching?
A spatial segmentation strategy for immersive media where only the specific tiles within a user's current viewport are pre-fetched and cached at high resolution, optimizing bandwidth for 360-degree video and extended reality applications.
The strategy relies on viewport prediction algorithms that forecast head movement trajectories using historical navigation data and content saliency maps. By integrating with Multi-access Edge Computing (MEC) platforms, tile-based caches can execute sub-second prefetching decisions, ensuring that when a user rotates their head, the high-quality tiles for the new viewport are already locally available. This approach directly addresses the motion-to-photon latency requirements of extended reality, preventing black borders and maintaining an immersive quality of experience without requiring the bandwidth of a full 360-degree stream.
Key Characteristics of Tile-Based Caching
Tile-based caching is a viewport-dependent strategy that decomposes 360-degree or high-resolution immersive media into discrete spatial segments, caching only the tiles within a user's current and predicted field of view at high resolution to dramatically reduce bandwidth consumption.
Viewport-Adaptive Streaming
The core mechanism of tile-based delivery, where only the tiles within the user's current field of view (FoV) are streamed at full resolution. Tiles outside the viewport are either not delivered or fetched at a significantly lower quality. This is governed by the HTTP Adaptive Streaming (HAS) paradigm, where a manifest file describes the spatial relationship of each tile segment. The client continuously reports its head orientation, and the edge server responds by pushing the relevant high-quality tiles, reducing total bitrate by up to 80% compared to full-panorama delivery.
Tiling Schemes and Encoding
The spatial partitioning of a projection map (e.g., Equirectangular Projection (ERP) or Cubemap) into a grid of independent, decodable bitstreams. Common schemes include fixed grids (e.g., 12x6 tiles) or adaptive grids with finer granularity at the equator. Each tile is encoded at multiple quality representations using standards like HEVC/H.265 tiles or VVC/H.266 subpictures, which enforce spatial independence by breaking prediction dependencies at tile boundaries. This allows the client to assemble a composite frame from tiles of varying resolutions without decoding artifacts at the seams.
Gaze-Driven Pre-fetching
A predictive caching layer that anticipates head movement to pre-fetch tiles before they enter the viewport. By analyzing angular velocity and saliency maps (regions statistically likely to attract attention), a machine learning model can predict the future viewport by 200-500ms. This margin is critical to mask network round-trip time. The system pre-fetches a probabilistic set of tiles into a MEC edge cache, ensuring that when a user turns their head, the high-resolution data is already local. This transforms caching from a reactive buffer to a predictive, latency-hiding mechanism.
Rate-Distortion Optimization
The mathematical framework for allocating bits across tiles to maximize perceived quality under a bandwidth constraint. The optimization function assigns higher bitrates to tiles within the foveated region (the center of the gaze) and lower bitrates to peripheral tiles. This exploits the natural fall-off in human visual acuity. The process solves a constrained optimization problem: minimize distortion (D) subject to a rate (R) budget, often using Lagrangian optimization to find the convex hull of the rate-distortion curve. The result is a non-uniform bitrate allocation that is perceptually indistinguishable from a uniformly high-quality stream.
Tile Boundary Artifact Mitigation
Techniques to eliminate visible seams when decoding tiles of different resolutions. Since independent encoding breaks in-loop filters like deblocking and sample adaptive offset (SAO) at tile edges, stitching mismatched tiles can create perceptual artifacts. Mitigation strategies include encoding with a guard band (overlapping pixel margin) that is cropped during rendering, or using motion-constrained tile sets (MCTS) that ensure temporal prediction does not reference samples across tile boundaries. The DASH-SRD (Spatial Relationship Description) standard extends MPEG-DASH to signal these spatial relationships to the client for seamless reconstruction.
Edge-Compliant Tile Packaging
The storage and delivery format optimized for low-latency edge caches. Instead of caching monolithic video files, the content is packaged as ISO Base Media File Format (ISOBMFF) segments where each tile is an independent track. A Common Media Application Format (CMAF) chunked encoding allows low-latency delivery by enabling the edge server to push individual CMAF chunks for a tile as soon as they are encoded, without waiting for a full segment. This chunked transfer encoding over HTTP/2 or HTTP/3 minimizes server-side buffering and allows the edge cache to act as a real-time relay.
Frequently Asked Questions
Clear, technically precise answers to the most common questions about tile-based caching for immersive media delivery, viewport-adaptive streaming, and edge pre-fetching strategies.
Tile-based caching is a spatial data management strategy for immersive media where a 360-degree video frame is segmented into independent, rectangular spatial sub-units called tiles. Instead of caching an entire monolithic video file, the system pre-fetches and stores only the specific tiles that fall within a user's predicted viewport at high resolution, while the remaining tiles are either not cached, stored at a lower quality, or fetched on demand. This process relies on viewport prediction algorithms—often using historical head movement data and sequence-aware recommendation models—to determine which tiles will be needed next. The cached tiles are placed at the Multi-access Edge Computing (MEC) node or a local base station cache, ensuring ultra-low latency delivery. When the user rotates their head, the client requests the new viewport tiles, and the cache eviction policy (such as LRU-K) removes tiles that are no longer spatially relevant, optimizing the cache hit ratio for the current field of view.
Enabling Efficiency, Speed & Accuracy
Intelligent Analysis, Decision & Execution
We build AI systems for teams that need search across company data, workflow automation across tools, or AI features inside products and internal software.
Talk to Us
Search across company data
Give teams answers from docs, tickets, runbooks, and product data with sources and permissions.
Useful when people spend too long searching or get different answers from different systems.

Automate internal workflows
Use AI to route work, draft outputs, trigger actions, and keep approvals and logs in place.
Useful when repetitive work moves across multiple tools and teams.

Add AI to products and internal tools
Build assistants, guided actions, or decision support into the software your team or customers already use.
Useful when AI needs to be part of the product, not a separate tool.
Related Terms
Master the core concepts that enable efficient, viewport-adaptive streaming of 360-degree and volumetric content.
Viewport-Adaptive Streaming
The foundational delivery paradigm where only the user's current field of view is streamed at the highest quality. This technique reduces total bandwidth consumption by 50-80% compared to delivering the full 360-degree sphere. It relies on viewport prediction algorithms that forecast head movement 1-2 seconds into the future to pre-buffer the correct tiles, preventing black patches during fast rotations.
DASH-SRD & HTTP Live Streaming (HLS) Tile
Standardized manifest protocols for tile-based delivery. MPEG-DASH SRD (Spatial Relationship Description) extends the Media Presentation Description to signal the spatial position and size of each tile within the 360-degree sphere. Apple's HLS uses IMSC1-style captions and tile playlists. These manifests allow the client player to dynamically select tile representations based on the current viewport and available bandwidth.
Equirectangular Projection (ERP) & Cubemap
The geometric mappings that project a spherical video onto a 2D plane for encoding. ERP is the most common format but introduces heavy pixel redundancy at the poles. Cubemap projection maps the sphere onto six square faces, reducing the peak sample density variation. Efficient tile-based caching must account for the projection format, as a user's viewport will span different tile regions depending on the mapping used.
Rate-Distortion Optimization for Tiles
The encoding strategy that allocates bitrate asymmetrically across the tile grid. Tiles statistically likely to fall within the user's initial viewport (e.g., the center of the action) are encoded at a higher PSNR quality, while periphery tiles are heavily quantized. This is often guided by a saliency map or content popularity heatmap, ensuring perceptual quality is maximized for the most-viewed spatial regions without increasing the total bitrate.

About the author
Prasad Kumkar
CEO & MD, Inference Systems
Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.
His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.
Partnered with leading AI, data, and software stack.
How We Work
Custom AI workflows for your Business
One-fit-all AI don't work for modern businesses. At Inferensys, we aim to understand your business & custom requirements; which we use to define most efficient agentic workflows, the data, and the tools for your business.
01
Review the use case
We understand the task, the users, and where AI can actually help.
Read more02
Pick the right approach
We define what needs search, automation, or product integration.
Read more03
Build the first useful version
We implement the part that proves the value first.
Read more04
Improve from there
We add the checks and visibility needed to keep it useful.
Read moreThe first call is a practical review of your use case and the right next step.
Talk to Us