Generation Is Compression: Zero-Shot Video Coding via Stochastic Rectified Flow

Dimensions overall score 7.0

• Generative Video Codec (GVC) is a zero-shot framework that transforms a pretrained video generative model into a video codec without retraining.method 95%
  Evidencepartial

  We propose Generative Video Codec (GVC), a zero-shot framework that turns a pretrained video generative model into the codec itself: the transmitted bitstream directly specifies the generative decoding trajectory, with no retraining required.

  Implicationpartial

  This is a core claim stated directly in the abstract and title, and elaborated upon in the introduction.

  Verificationpartial

  partial

• GVC converts the deterministic rectified-flow ODE of video foundation models into an equivalent SDE at inference time for codebook-driven compression.method 95%
  Evidencepartial

  To enable this, we convert the deterministic rectified-flow ODE of modern video foundation models into an equivalent SDE at inference time, unlocking per-step stochastic injection points for codebook-driven compression.

  Implicationpartial

  This is a key technical innovation described in the abstract and detailed in the methods section.

  Verificationpartial

  partial

• GVC achieves high-quality reconstruction below 0.002 bpp.result 90%
  Evidencepartial

  Experiments on standard benchmarks show that GVC achieves high-quality reconstruction below 0.002 bpp while supporting flexible bitrate control through a single hyperparameter.

  Implicationpartial

  This is a specific quantitative result presented in the abstract and supported by experimental results tables.

  Verificationpartial

  partial

• GVC is preferred over DCVC-RT in 97% of pairwise comparisons in a user study.result 90%
  Evidencepartial

  a user study confirms that GVC is preferred over DCVC-RT in 97% and over GNVC-VD in 88% of pairwise comparisons.

  Implicationpartial

  This is a strong user study result directly comparing GVC to a baseline.

  Verificationpartial

  partial

• The scalar gscale serves as the primary knob for the rate–quality trade-off in GVC.technical 90%
  Evidencepartial

  The scalar gscale serves as the primary knob for the rate–quality trade-off.

  Implicationpartial

  This describes a key parameter controlling the performance trade-off, as explained in the methods section.

  Verificationpartial

  partial

• The Text-to-Video (T2V) variant of GVC operates at near-zero side information as a pure generative prior.method 90%
  Evidencepartial

  Text-to-Video (T2V) operating at near-zero side information as a pure generative prior

  Implicationpartial

  This describes a specific variant and its operational characteristic, as stated in the abstract and detailed in the methods.

  Verificationpartial

  partial

• A limitation of T2V is the absence of spatial anchoring, which can lead to subtle positional drift or content deviation across GOPs.limitation 85%
  Evidencepartial

  A known limitation of T2V is the absence of spatial anchoring: without a reference frame, the model may produce subtle positional drift or content deviation across GOPs.

  Implicationpartial

  This identifies a specific limitation of one of the proposed variants, as discussed in the text.

  Verificationpartial

  partial

• GVC recovers sharp textures and coherent details while DCVC-RT exhibits severe oversmoothing at comparable bitrates.result 90%
  Evidencepartial

  GVC recovers sharp textures and coherent details while DCVC-RT exhibits severe oversmoothing.

  Implicationpartial

  This is a qualitative result comparing GVC to a baseline, supported by visual examples and descriptions.

  Verificationpartial

  partial

• Generative Video Codec (GVC) is a zero-shot framework that transforms a pretrained video generative model into a video codec without retraining.method 95%
  Evidencepartial

  We propose Generative Video Codec(GVC), a zero-shot framework that turns a pretrained video generative model into the codec itself: the transmitted bitstream directly specifies the generative decoding trajectory, with no retraining required.

  Implicationpartial

  This is a core claim stated directly in the abstract and introduction.

  Verificationpartial

  partial

• GVC converts the deterministic rectified-flow ODE of video foundation models into an equivalent SDE at inference time for codebook-driven compression.method 95%
  Evidencepartial

  To enable this, we convert the deterministic rectified-flow ODE of modern video foundation models into an equivalent SDE at inference time, unlocking per-step stochastic injection points for codebook-driven compression.

  Implicationpartial

  This is a key technical innovation described in the abstract and detailed in the methods section.

  Verificationpartial

  partial

• GVC achieves high-quality reconstruction below 0.002 bits per pixel (bpp).result 90%
  Evidencepartial

  Experiments on standard benchmarks show that GVC achieves high-quality reconstruction below 0.002\,bpp while supporting flexible bitrate control through a single hyperparameter.

  Implicationpartial

  This is a specific quantitative result presented in the abstract and supported by tables in the paper.

  Verificationpartial

  partial

• GVC achieves a 70.3% reduction in LPIPS over DCVC-RT at comparable bitrates.result 90%
  Evidencepartial

  LPIPS comparison shows GVC achieves a 70.3% reduction over DCVC-RT at comparable bitrate; a user study confirms that GVC is preferred over DCVC-RT in 97% and over GNVC-VD in 88% of pairwise comparisons.

  Implicationpartial

  This is a specific quantitative comparison result provided in the text.

  Verificationpartial

  partial

Competitive landscape

A zero-shot video codec that leverages pretrained generative models for high-quality, flexible bitrate video compression.