MemCam: Memory-Augmented Camera Control for Consistent Video Generation

Dimensions overall score 5.0

• MemCam significantly outperforms existing baseline methods and state-of-the-art approaches in terms of scene consistency, especially in long video generation with large camera rotations.result 95%
  Evidencepartial

  MemCam achieves competitive or superior performance acr

  Implicationpartial

  The abstract explicitly states this claim, and Table I provides quantitative results showing MemCam achieving the best FVD scores across different scenarios, indicating superior temporal consistency.

  Verificationpartial

  partial

• MemCam achieves the best FVD (Fréchet Video Distance) across all tested settings, signifying superior temporal consistency and visual quality.result 95%
  Evidencepartial

  MemCam achieves the best FVD across all settings, indicating superior temporal consistency and visual quality.

  Implicationpartial

  The text directly states that MemCam achieves the best FVD and explains what this metric indicates. Table I provides the numerical FVD values supporting this.

  Verificationpartial

  partial

• The proposed context compression module effectively reduces computational overhead while enriching contextual information for video generation.method 90%
  Evidencepartial

  thereby reducing computational overhead while enriching contextual information.

  Implicationpartial

  The abstract mentions the context compression module's purpose is to reduce computational overhead and enrich context. Table III shows a reduction in 's/frame' for MemCam variants compared to 'None' variants, supporting the overhead reduction aspect.

  Verificationpartial

  partial

• MemCam faithfully preserves scene structure even after large camera rotations, outperforming other methods that exhibit scene inconsistency.result 90%
  Evidencepartial

  As shown in Fig. 3, MemCam faithfully preserves the scene structure even after large camera rotations, while other methods exhibit

  Implicationpartial

  Figure 3 visually demonstrates this claim, and the accompanying text explicitly states MemCam's superior performance in preserving scene structure compared to other methods.

  Verificationpartial

  partial

• The co-visibility-based selection strategy for retrieving historical frames is crucial for memory retention and generation quality.method 85%
  Evidencepartial

  The results show that effective utilization of historical frames through compression and retrieval is crucial for memory retention and generation quality.

  Implicationpartial

  The abstract mentions co-visibility-based selection as part of the approach. Table II shows that the 'Ours' strategy (which includes co-visibility) outperforms 'Recent', 'Random', and 'TopK' in FVD and other metrics, supporting its importance.

  Verificationpartial

  partial

• Geometry-Forcing (GF) method's performance drops notably in 360° scenarios due to accumulated reconstruction errors.limitation 85%
  Evidencepartial

  GF achieves competitive PSNR/SSIM on 90° tasks, but its performance drops notably in 360° scenarios as reconstruction errors accumulate.

  Implicationpartial

  The text explicitly states this limitation of the GF method, and Table I shows GF's FVD score for 360° round-trip is significantly higher than MemCam's.

  Verificationpartial

  partial

• MemCam is built upon the Wan2.1 1.3B Text-to-Video Diffusion Transformer architecture.technical 95%
  Evidencepartial

  Our method is built on the Wan2.1 1.3B Text-to-Video Diffusion Transformer [1].

  Implicationpartial

  This is a direct statement about the implementation details of MemCam.

  Verificationpartial

  partial

Competitive landscape

A memory-augmented approach for consistent interactive video generation that maintains scene coherence under dynamic camera control.