Compression Efficiency

1. AV2achieves 28.6% PSNR YUV BD rate savings and 326% VMAF bit rate savings compared to AV1 in random access configuration.

2. AV2uses a hybrid block-based model with larger super blocks up to 256x256, recursive partitioning, and semi-decoupled partitioning for improved compression.

Advanced Prediction Techniques

3. AV2utilizes up to seven reference pictures arranged by temporal distance, quantizer, and frame resolution, with compound prediction modes including weighted prediction and temporal interpolated prediction.

4. Block adaptive weighted prediction for illumination compensation, motion trajectory tracking with nonlinear motion models, and reference MV bank and warp parameters bank enhance prediction accuracy.

Transform and Filtering Improvements

5. AV2implements transform block partitioning with one level of transform split, different partitions for luma and chroma, and transform improvements using matrix multiplications and learned unitary transforms.

6. Loop filters with generalized formula for multiple filter lengths, improved deblocking, cross component sample offsets, and guided detail filter improve picture quality and reduce bit rate.