AcrossVAM2.0: Conflict-Aware Point Supervision for Parameter-Efficient Video Diffusion Adaptation

arXiv cs.AI Papers
video-diffusion parameter-efficient-fine-tuning lora point-tracking optical-flow diffusion-transformer

Summary

AcrossVAM2.0 提出一种冲突感知的点对应监督方法，用于参数高效的视频扩散模型适配，通过置信度加权跟踪头、零初始化残差反馈和非对称 PCGrad，在保持预训练外观先验的同时提升对应一致性，但并不均匀改善视频质量。

arXiv:2610.14693v1 Announce Type: new Abstract: Explicit point correspondence is attractive for adapting video generators: it describes how visible surfaces move, but can overwrite the appearance prior learned by a large pretrained model. We study this tension in a 5.01B-parameter text-to-video diffusion transformer. ACROSSVAM2.0 combines a confidence-weighted track-and-visibility head, zero-initialized residual feedback, and asymmetric PCGrad that removes only the auxiliary component opposing the primary flow-matching gradient on shared LoRA parameters. The design preserves the generator exactly at initialization and protects its primary update during adaptation. In a four-way ablation trained for 1,000 steps with three independent seeds, the full method lowers KLT forward-backward cycle error in all three seeds relative to a matched flow-only continuation (mean seed-level difference -0.00711, SD 0.00505), while robust sharpness remains positive in all three. An expanded 24-prompt evaluation (216 videos) yields 9.7% lower aggregate CoTracker3 cycle error than the matched control and near-matched performance on six VBench dimensions; however, the seed-level tracking effect is heterogeneous and I3D-FVD regresses from 0.1824 to 0.2022. These results support a bounded claim: conflict-aware adaptation improves correspondence consistency without broad perceptual degradation, but does not uniformly improve video quality. We release equations, pseudocode, matched outputs, source tables, and artifact provenance.
Original Article
View Cached Full Text

Cached at: 10/15/26, 06:55 AM 

# AcrossVAM2.0: Conflict-Aware Point Supervision for Parameter-Efficient Video Diffusion Adaptation
Source: [https://arxiv.org/abs/2610.14693](https://arxiv.org/abs/2610.14693)
[View PDF](https://arxiv.org/pdf/2610.14693)

> Abstract:Explicit point correspondence is attractive for adapting video generators: it describes how visible surfaces move, but can overwrite the appearance prior learned by a large pretrained model\. We study this tension in a 5\.01B\-parameter text\-to\-video diffusion transformer\. ACROSSVAM2\.0 combines a confidence\-weighted track\-and\-visibility head, zero\-initialized residual feedback, and asymmetric PCGrad that removes only the auxiliary component opposing the primary flow\-matching gradient on shared LoRA parameters\. The design preserves the generator exactly at initialization and protects its primary update during adaptation\. In a four\-way ablation trained for 1,000 steps with three independent seeds, the full method lowers KLT forward\-backward cycle error in all three seeds relative to a matched flow\-only continuation \(mean seed\-level difference \-0\.00711, SD 0\.00505\), while robust sharpness remains positive in all three\. An expanded 24\-prompt evaluation \(216 videos\) yields 9\.7% lower aggregate CoTracker3 cycle error than the matched control and near\-matched performance on six VBench dimensions; however, the seed\-level tracking effect is heterogeneous and I3D\-FVD regresses from 0\.1824 to 0\.2022\. These results support a bounded claim: conflict\-aware adaptation improves correspondence consistency without broad perceptual degradation, but does not uniformly improve video quality\. We release equations, pseudocode, matched outputs, source tables, and artifact provenance\.

## Submission history

From: Nan Wu \[[view email](https://arxiv.org/show-email/8edcfecf/2610.14693)\] **\[v1\]**Sat, 10 Oct 2026 08:12:08 UTC \(4,164 KB\)

Similar Articles