1/8-Pel Motion Vectors in AV1 Motion Compensation

This article explores how 1/8th-pel (eighth-pixel) motion vector precision functions within the AV1 video codec to enhance motion compensation. By allowing the codec to track motion at fractional pixel increments twice as fine as the 1/4-pel standard used in previous codecs, AV1 significantly reduces prediction error, preserves fine textures during slow or subtle motion, and decreases the amount of residual data that must be compressed into the final bitstream.

The Mechanics of Sub-Pixel Precision

In video compression, inter-frame prediction uses motion vectors to reference pixel blocks from previous frames to describe the current frame. Real-world motion is continuous and rarely aligns perfectly with the discrete pixel grid of a camera sensor. To capture displacement that falls between pixels, codecs calculate fractional positions using interpolation.

While legacy standards like H.264 and HEVC predominantly rely on 1/4-pel (quarter-pixel) accuracy, AV1 introduces native support for 1/8-pel accuracy. This divides the distance between any two pixels into eight discrete fractional steps rather than four, allowing the motion estimator to select reference blocks that match the target frame far more accurately.

Minimizing the Residual Error

The primary goal of motion compensation is to minimize the "residual"—the difference between the predicted block and the actual block. The smaller the residual error, the fewer bits the transform and quantization stages require to represent the frame.

When motion shifts an object by a fraction that does not fall cleanly on a quarter-pixel boundary (such as 0.375 or 0.625 of a pixel), a 1/4-pel system must round to the nearest available step. This slight misalignment introduces high-frequency error into the prediction block. By using 1/8th-pel precision, AV1 cuts this rounding error in half. The reference block matches the source block much more closely, which directly translates to a lower-energy residual signal and substantial bitrate savings.

Improved Detail Preservation via Advanced Interpolation

To compute 1/8-pel positions, AV1 employs high-precision, separable multi-tap interpolation filters (up to 8-tap filters for luminance). These filters reconstruct intermediate sub-pixel color and brightness values smoothly while suppressing aliasing.

With 1/4-pel interpolation, the broader steps can introduce minor blurring across sequential frames, especially during slow, continuous movement. AV1's 1/8th-pel precision, paired with sharp interpolation filters, prevents this blurring. Fine edges, complex textures, and high-frequency details (such as text, grain, or foliage) remain intact across frames without bleeding into neighboring pixels.

Impact on High-Resolution and Low-Velocity Motion

The advantages of 1/8th-pel accuracy become especially pronounced under specific conditions:

• High-Resolution Content (4K and 8K): In ultra-high-definition video, a single pixel represents a much smaller physical angle in the field of view. Real-world movements translate into tiny, sub-pixel shifts on the sensor grid, making sub-pixel granularity critical for effective tracking.
• Slow Camera Pans and Zooms: Gentle camera movements often move image features by fractions of a pixel per frame. 1/8th-pel accuracy prevents the visual "jittering" or blocky artifacts that occur when motion vectors are forced to snap to larger fractional intervals.
• Low-Bitrate Targets: When bandwidth is constrained, codecs discard high-frequency residual information. By ensuring the motion prediction is naturally accurate down to an eighth of a pixel, the codec relies less on the residual data to maintain image clarity.

By refining motion estimation to the 1/8th-pixel level, AV1 achieves a tighter correlation between frames, lower residual data overhead, and superior visual stability in moving scenes.