Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cross-Platform Neural Video Coding: A Case Study

Ruhan Conceição, Marcelo Porto, Wen-Hsiao Peng, Luciano Agostini

TL;DR

The static quantization of the hyper prior decoding path is proposed, which effectively mitigates the mismatch issue and enhances compression efficiency results by preventing severe image quality degradation.

Abstract

In this paper, we first show that current learning-based video codecs, specifically the SSF codec, are not suitable for real-world applications due to the mismatch between the encoder and decoder caused by floating-point round-off errors. To address this issue, we propose the static quantization of the hyper prior decoding path. The quantization parameters are determined through an exhaustive search of all possible combinations of observers and quantization schemes from PyTorch. For the SSF codec, when encoding and decoding on different machines, the proposed solution effectively mitigates the mismatch issue and enhances compression efficiency results by preventing severe image quality degradation. When encoding and decoding are performed on the same machine, it constrains the average BD-rate increase to 9.93% and 9.02% for UVG and HEVC-B sequences, respectively.

Cross-Platform Neural Video Coding: A Case Study

TL;DR

The static quantization of the hyper prior decoding path is proposed, which effectively mitigates the mismatch issue and enhances compression efficiency results by preventing severe image quality degradation.

Abstract

In this paper, we first show that current learning-based video codecs, specifically the SSF codec, are not suitable for real-world applications due to the mismatch between the encoder and decoder caused by floating-point round-off errors. To address this issue, we propose the static quantization of the hyper prior decoding path. The quantization parameters are determined through an exhaustive search of all possible combinations of observers and quantization schemes from PyTorch. For the SSF codec, when encoding and decoding on different machines, the proposed solution effectively mitigates the mismatch issue and enhances compression efficiency results by preventing severe image quality degradation. When encoding and decoding are performed on the same machine, it constrains the average BD-rate increase to 9.93% and 9.02% for UVG and HEVC-B sequences, respectively.

Paper Structure

This paper contains 13 sections, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background
  3. The SSF Codec
  4. Literature Review on Cross-platform Mismatch Mitigation
  5. Quantization
  6. Extra Signaling
  7. PyTorch Built-in Quantization Framework
  8. Floating-point Round-off Errors
  9. Static Hyper Prior Decoding Quantization
  10. Quantization Parameters Exhaustive Search
  11. Exhaustive Search Results
  12. Experimental Results
  13. Concluding Remarks

Figures (5)

  • Figure 1: Floating-point error propagation across different SSF quality levels. Full-resolution images are available at https://ssfquant-iscas2025.netlify.app/
  • Figure 2: SSF Codec Block Diagram
  • Figure 3: Rate-distortion curves for: a) all quantization configurations; b) configurations with a BD-rate increase of less than 30%.
  • Figure 4: Mitigation of floating-point error propagation across different SSF quality levels due to hyper prior decoding path quantization.
  • Figure 5: PSNR curves for four scenarios comparing quantized and non-quantized SSF versions when decoding the bitstream generated on same and different machines.