AV1 Tile Grids for Equirectangular 360 Video

This article examines how the equirectangular projection (ERP) format interfaces with the AV1 video codec's tile grid architecture to deliver high-resolution 360-degree video. By dividing the distorted two-dimensional ERP canvas into independently decodable rectangular segments, AV1 tile grids facilitate viewport-adaptive streaming, reduce computational bottlenecks, and allow targeted compression adjustments that counteract the uneven pixel density typical of spherical video mapping.

The Equirectangular Projection Canvas

Equirectangular projection maps the surface of a sphere onto a flat, two-dimensional rectangular surface with a standard 2:1 aspect ratio. The horizontal coordinate represents longitude, while the vertical coordinate represents latitude.

Because the projection stretches the spherical poles across the full width of the top and bottom edges, it creates extreme geometric distortion. The pixel density near the poles is artificially inflated compared to the equator, resulting in redundant visual information and heavy data footprints for areas viewers rarely focus on.

AV1 Tile Grid Mechanics

The AV1 codec features an advanced spatial partitioning tool known as tile grids. A frame can be subdivided into horizontal rows and vertical columns of rectangular blocks called tiles.

AV1 supports both uniform and non-uniform tile spacing:

• Uniform spacing: Divides the frame into equally sized grid blocks based on power-of-two calculations.
• Non-uniform spacing: Allows custom configurations where tile widths and heights vary across the frame layout.

Crucially, AV1 allows tiles to be encoded and decoded independently. Spatial intra-prediction does not cross tile boundaries when configured for independent operation, and loop filtering across tile edges can be disabled or handled deterministically. Each tile or collection of tiles can be packaged into individual Tile Group Open Bitstream Units (OBUs).

Interfacing ERP with AV1 Tiles

The interface between ERP and AV1 tile grids relies on spatially isolating spherical coordinate zones into independent encoded structures.

1. Viewport-Adaptive Streaming

At any given moment in an immersive 360-degree experience, a user's field of view (FoV) covers only about 90 to 110 degrees of the full spherical surface. Transmitting an entire 8K ERP video at uniform high quality wastes substantial network bandwidth.

By aligning an AV1 tile grid to the ERP canvas, streaming engines can implement tiled viewport-dependent delivery:

• The client detects the user’s head orientation.
• The video player fetches high-bitrate AV1 tiles covering the primary visual field.
• The remaining tiles covering the peripheral and rear regions of the ERP frame are either fetched at a significantly lower bitrate or omitted entirely.
• The AV1 decoder reconstructs the selected tile configuration to maintain a seamless visual environment without overloading the network pipeline.

2. Distortion Compensation via Dynamic Bitrate Allocation

Because ERP exhibits varying degrees of distortion depending on the latitude, AV1 tiles allow encoders to treat regions differently based on their vertical position on the sphere:

• Equatorial Tiles: These tiles carry the most critical visual data with minimal distortion and typically attract the viewer’s primary attention. Encoders assign lower Quantization Parameters (QP) to these central tiles to maximize fidelity.
• Polar Tiles: The upper and lower rows of the tile grid correspond to the stretched polar regions. Encoders can assign higher QP values or apply aggressive spatial filtering to these tiles, suppressing the high data rates typically demanded by the stretched, oversampled pixels.

3. Asymmetric Grid Layouts

Using AV1's non-uniform tile sizing, video architects can design grids tailored specifically to ERP characteristics. For instance, an encoder can create narrower vertical bands near the center of the frame where viewport transitions require fine-grained data fetching, and larger, broader tiles at the extreme top and bottom where viewer interaction is low and structural precision is less critical.

Decoding Efficiency and Hardware Benefits

High-resolution ERP content (such as 6K or 8K) demands immense memory bandwidth and processing power. AV1 tile grids enable multi-threaded parallel decoding, where different CPU or GPU cores simultaneously decode separate tiles of the projected sphere before passing them to the rendering engine for spherical projection mapping. This parallelism is essential for mobile virtual reality headsets and low-power devices that require low-latency decoding to prevent motion sickness during rapid head movements.