Shape-of-You: Fused Gromov-Wasserstein Optimal Transport for Semantic Correspondence in-the-Wild

Dimensions overall score 8.0

• The Shape-of-You (SoY) framework reformulates pseudo-label generation for semantic correspondence as a Fused Gromov-Wasserstein (FGW) problem.method 95%
  Evidencepartial

  In this work, we address this by reformulating pseudo-label generation as a Fused Gromov-Wasserstein (FGW) problem

  Implicationpartial

  Directly and explicitly stated in the abstract as the core methodological contribution.

  Verificationpartial

  partial

• The FGW formulation jointly optimizes inter-feature similarity and intra-structural consistency for semantic correspondence.method 90%
  Evidencepartial

  which jointly optimizes inter-feature similarity and intra-structural consistency.

  Implicationpartial

  Directly stated in the abstract, clearly defining the two key terms optimized by the proposed method.

  Verificationpartial

  partial

• Shape-of-You leverages a 3D foundation model to define intra-structure in geometric space to resolve geometric ambiguities.method 90%
  Evidencepartial

  Our framework, Shape-of-You (SoY), leverages a 3D foundation model to define this intra-structure in the geometric space, resolving abovementioned ambiguity.

  Implicationpartial

  Directly stated in the abstract as a key component of the method and its purpose.

  Verificationpartial

  partial

• The FGW problem is computationally prohibitive, so SoY approximates it through anchor-based linearization.technical 85%
  Evidencepartial

  However, since FGW is a computationally prohibitive quadratic problem, we approximate it through anchor-based linearization.

  Implicationpartial

  Directly stated in the abstract as a challenge and the specific solution implemented.

  Verificationpartial

  partial

• SoY introduces a soft-target loss that dynamically blends guidance from the FGW transport plan with network predictions to handle noise.method 85%
  Evidencepartial

  Thus, we introduce a soft-target loss dynamically blending guidance from this plan with network predictions to build a learning framework robust to this noise.

  Implicationpartial

  Directly stated in the abstract as a specific technical contribution to address a stated problem.

  Verificationpartial

  partial

• Shape-of-You achieves state-of-the-art performance on the SPair-71k and AP-10k datasets for semantic correspondence.result 95%
  Evidencepartial

  SoY achieves state-of-the-art performance on SPair-71k and AP-10k datasets

  Implicationpartial

  Explicitly and directly stated as a key result in the abstract.

  Verificationpartial

  partial

• The SoY framework establishes a new benchmark in semantic correspondence without using explicit geometric annotations.result 90%
  Evidencepartial

  establishing a new benchmark in semantic correspondence without explicit geometric annotations.

  Implicationpartial

  Directly stated in the abstract as a summary of the achievement.

  Verificationpartial

  partial

• A key limitation of prior nearest-neighbor pseudo-label methods is their local operation and failure to resolve geometric ambiguities from symmetries or repetitive features.limitation 85%
  Evidencepartial

  it operates locally, ignoring structural relationships, and consequently its reliance on 2D appearance fails to resolve geometric ambiguities arising from symmetries or repetitive features.

  Implicationpartial

  Directly stated in the abstract as the problem motivation, though framed as a limitation of prior work.

  Verificationpartial

  partial

Competitive landscape

Shape-of-You offers a novel approach to semantic correspondence using Fused Gromov-Wasserstein optimal transport, achieving state-of-the-art results without explicit geometric annotations.