Oct 21, 2026 · Artificial Intelligence

MoFu: Solving Scale & Order Challenges in Multi-Subject Video Generation

MoFu introduces scale-aware modulation and Fourier fusion to fix scale inconsistency and permutation sensitivity in multi-subject video generation, adds a joint stability loss and benchmark, and outperforms SOTA methods on scale, order robustness, and overall quality.

AAAI 2026Diffusion ModelsFourier fusion

0 likes · 15 min read

MoFu: Solving Scale & Order Challenges in Multi-Subject Video Generation

Alibaba Cloud Big Data AI Platform

Oct 8, 2023 · Artificial Intelligence

Why the Scale‑Aware Modulation Transformer Outperforms CNNs and Vision Transformers with Fewer Parameters

The Scale‑Aware Modulation Transformer (SMT) introduces a lightweight SAM module and an Evolutionary Hybrid Network that together achieve higher accuracy on ImageNet, COCO, and ADE20K while using significantly fewer parameters and FLOPs than existing CNN and Transformer baselines.

Object DetectionSMThybrid CNN‑Transformer

0 likes · 12 min read

Why the Scale‑Aware Modulation Transformer Outperforms CNNs and Vision Transformers with Fewer Parameters