Current Issue

In VVC/H.266, ISP tool enhances compression efficiency by dividing a coding block into smaller sub-blocks, allowing each sub-block to be predicted and encoded separately. This approach is especially beneficial for complex textures and detailed areas, as it enables more accurate intra predictions and reduces prediction errors.

As shown in Table 2, the performance changes by 0.52%, -0.36%, and 0.06% for the Y, Cb, and Cr components on average, compared to the default settings. The tool-off testing results suggest that the impact of disabling ISP is relatively comparable to the results observed in natural camera-captured videos^[16]. This consistency highlights that ISP’s contribution to compression efficiency in gaming content is similar to its role in conventional video content.

MRL tool in VVC/H.266 enhances intra prediction by utilizing multiple reference lines from neighboring blocks instead of relying on one. This approach enables more precise predictions by capturing a wider range of directional patterns, which is particularly beneficial for encoding complex textures. By improving prediction accuracy, MRL helps reduce residual errors, leading to better overall compression efficiency.

As shown in Table 3, the performance changes by 0.30%, 1.85%, and 0.98% for the Y, Cb, and Cr components on average, compared to the default settings. Overall, the tool-off test results indicate that the impact of disabling MRL is quite similar to what is observed in natural camera-captured videos^[16]. Notably, MRL has a significant impact on the Minecraft sequence, suggesting that the use of multiple reference lines for intra prediction is highly beneficial even in gaming content, where additional directional references can greatly enhance compression efficiency.

MIP tool in VVC/H.266 introduces an innovative approach to intra prediction by using matrix-based transformations to predict pixel values. This method is particularly effective for blocks with non-linear or complex textures, where conventional directional prediction methods may be less accurate. MIP enhances prediction precision in such cases, leading to improved compression efficiency.

As shown in Table 4, the performance changes by 0.64%, 2.04%, and 0.99% for the Y, Cb, and Cr components on average, compared to the default settings. The tool-off test results indicate that the impact of disabling MIP is similar to that observed in natural camera-captured videos, with a slight improvement in Chroma coding performance^[16].

IBC tool in VVC/H.266 is designed for compressing screen content by reusing previously reconstructed blocks within the same frame. It functions similarly to inter prediction but operates within the same frame, allowing for efficient encoding of repeating patterns, text, or graphics. By leveraging existing data within a frame, IBC significantly improves compression efficiency for screen content applications, reducing the need for redundant coding.

As shown in Table 5, the performance changes by -3.37%, -2.84%, and 0-2.98% for the Y, Cb, and Cr components on average, compared to the default settings. Since IBC is disabled by default in the AI configuration, we conducted a tool-on test to assess its impact. The results demonstrate a significant reduction in BD-rate, indicating that enabling IBC leads to notable improvements. Given that IBC is designed to provide accurate predictions for screen content, it is reasonable to observe such substantial performance gains in gaming sequences.

TS tool in VVC/H.266 allows certain blocks of residual data to bypass the transform stage, which normally converts spatial domain data to the frequency domain. This is particularly useful for small blocks or regions with simple textures, where applying a transform may not provide compression gains.

As shown in Table 6, the performance changes by 2.01%, 2.26%, and 1.38% for the Y, Cb, and Cr components on average, compared to the default settings. The tool-off test results show a significant reduction in BD-rate, particularly for the Minecraft sequence. Given that Transform Skip is designed to handle screen content where the prediction closely matches the original, it is reasonable to observe such notable improvements in gaming content.

MTS tool in VVC/H.266 enhances the compression process by offering a selection of different transform types including DCT-II and DST-VII to better adapt to varying block characteristics. By selecting the most suitable transform for each block, MTS improves the representation of both smooth and detailed areas, leading to more efficient residual compression.

As shown in Table 7, the performance changes by 0.83%, 1.18%, and 0.50% for the Y, Cb, and Cr components on average, compared to the default settings. The tool-off test results indicate that the impact of disabling MTS is similar to that observed in natural camera-captured videos^[16]. However, an interesting result was observed in the Minecraft sequence. This can be explained by the characteristics of Minecraft, which are more aligned with screen content, while the other sequences resemble natural content in their transform usage, leading to different results across sequences.

LFNST tool in VVC/H.266 improves compression efficiency by applying an additional transform to low-frequency components of intra-predicted blocks. This tool is designed to capture residual energy more effectively in smooth or flat regions, where traditional transforms may not provide optimal compression. By refining the representation of low-frequency details, LFNST helps reduce the overall bitrate while preserving video quality, particularly for high-resolution and low-texture areas.

As shown in Table 8, the performance changes by 0.45%, 1.40%, and 1.81% for the Y, Cb, and Cr components on average, compared to the default settings. The tool-off test results indicate that disabling LFNST has a smaller impact compared to natural camera-captured videos^[16]. This may be due to the more uniform residual patterns in gaming sequences, which differ from the more varied textures in natural content. As a result, the additional transform applied by LFNST is less effective in gaming sequences, where intra-prediction and core transforms are already highly efficient.

SAO tool, carried over from HEVC to VVC/H.266, enhances video quality by reducing artifacts and improving edge preservation after the primary compression process. It works by applying offsets to reconstructed pixels based on the characteristics of neighboring pixels, thus correcting distortion introduced during block-based compression. SAO effectively improves visual quality while maintaining compression efficiency, making it useful for both high-resolution and detailed content.

As shown in Table 9, the performance changes by 0.15%, 0.37%, and 0.19% for the Y, Cb, and Cr components on average, compared to the default settings. The tool-off test results indicate that the impact of disabling SAO is similar to what is observed in natural camera-captured videos^[16]. This outcome is expected, as SAO was primarily designed to enhance the visual quality of reconstructed video rather than to improve objective coding performance. However, the results suggest that SAO may be particularly beneficial for artificially generated content, as evidenced by the improved coding performance observed in the Minecraft sequence.

ALF tool in VVC/H.266 refines the quality of reconstructed video frames by applying a filter after the decoding process, targeting specific areas to reduce artifacts and improve visual quality. ALF dynamically adjusts its filtering based on the content of each frame, enhancing edge detail and reducing block artifacts introduced during compression. This tool helps improve overall video quality while maintaining compression efficiency, particularly for high-resolution and complex content.

As shown in Table 10, the performance changes by 1.90%, 14.06%, and 13.21% for the Y, Cb, and Cr components on average, compared to the default settings. The tool-off test results indicate that the impact of disabling ALF is smaller than what is observed in natural camera-captured videos^[16]. However, the coding improvement in the Chroma components is particularly noteworthy, demonstrating ALF’s effectiveness in enhancing color accuracy and compression efficiency.

LMCS tool in VVC/H.266 enhances video quality by adjusting the brightness (i.e., Luma) and color (i.e., Chroma) components separately to better match the visual characteristics of the content. It applies dynamic scaling to luma values, which allows for more efficient encoding of scenes with varying brightness levels, while simultaneously ensuring that the chroma components are adjusted accordingly.

As shown in Table 11, the performance changes by 0.15%, 0.02%, and 0.08% for the Y, Cb, and Cr components on average, compared to the default settings. The tool-off test results indicate that the impact of disabling LMCS is smaller than what is observed in natural camera-captured videos^[16]. Notably, the sequences in the G1 class show no coding improvements for either Luma or Chroma components, suggesting that the effectiveness of LMCS may vary depending on the characteristics of the content.

LMChroma tool in VVC/H.266 improves chroma (e.g., color) prediction accuracy by utilizing information from the luma (e.g., brightness) component to enhance chroma prediction. This tool exploits the correlation between luma and chroma components, adjusting chroma values based on the luma content within a block. By doing so, LMChroma enhances color representation and reduces residual errors, leading to better compression efficiency and visual quality, especially in scenes with complex color textures.

As shown in Table 12, the performance changes by 2.21%, 69.63%, and 42.25% for the Y, Cb, and Cr components on average, compared to the default settings. The tool-off test results demonstrate significant improvements in both Luma and Chroma components, with particularly impressive gains in Chroma performance. This suggests that the relationship between Luma and Chroma in gaming content is stronger than in natural camera-captured videos, potentially allowing these tools to achieve more effective compression in gaming sequences.

JCCR tool in VVC/H.266 enhances compression efficiency by jointly coding the chroma components (i.e., Cb and Cr), rather than treating them independently. This tool leverages the correlation between the two chroma channels, allowing for more efficient prediction and reduced residual data.

As shown in Table 13, the performance changes by 0.23%, 1.03%, and 0.34% for the Y, Cb, and Cr components on average, compared to the default settings. The tool-off test results indicate that the impact of disabling JCCR is smaller than what is typically observed in natural camera-captured videos^[16]. This suggests that the benefits of JCCR in gaming content may be less pronounced, possibly due to differences in the chroma-luma relationships between gaming sequences and natural video content.

ChromaTS tool in VVC/H.266 improves compression efficiency for chroma components by allowing certain chroma blocks to bypass the transform stage, similar to the transform skip for luma. This is particularly useful for chroma blocks with simple or flat color regions, where a transform may not offer significant gains. By skipping the transform in these cases, ChromaTS helps reduce unnecessary complexity and preserve color details, optimizing compression for specific content types.

As shown in Table 14, the performance changes by 0.00%, 1.59%, and 0.90% for the Y, Cb, and Cr components on average, compared to the default settings. The tool-off test results suggest that disabling ChromaTS has a relatively small impact on gaming content compared to TS tool. This indicates that the current design of ChromaTS may not align well with the characteristics of chroma components in gaming sequences, limiting its effectiveness in this context.