ACT Ⅲ
Large-Scale Dataset: ActCutVid-200k
Large-scale multi-shot video corpus with hierarchical cinematographic annotation.
Construction Pipeline
1
Source Collection
Large-scale video corpus from diverse genres, styles, and cinematographic traditions
2
AC Detection
SCBO algorithm identifies action-cut boundaries with boundary-residual optimization
3
Shot Segmentation
Audio-video segmentation into shot-level clips with precise boundary frame extraction
4
Hierarchical Annotation
Global scene + character captions; local static/dynamic cinematography; shot-relation labels
5
Quality Filtering
LLM-based annotation verification and expert review to ensure label accuracy and clip quality
Statistics
By Visual Style
Live Action / Realistic —%
2D Animation / Anime —%
3D CGI —%
Documentary / Non-Fiction —%
Other / Mixed —%
By Shot Count per Sequence
2-Shot —%
3-Shot —%
4-Shot —%
5-Shot —%
6+ Shot —%
By Action Complexity
Simple (Single action) —%
Medium (Action sequence) —%
Complex (Coordinated motion) —%
Out-of-Domain (OOD) —%
By Character Count
Single-Character —%
Two-Character —%
Multi-Character (3+) —%
Background Crowd —%
Sample Data
S1→S2
Live Action · 2-Shot · Simple Action
Match on Action: Hair-tying (ECU→MCU)
S1→S2
Anime · 3-Shot · Medium Action
Shot-Reverse-Shot: Dialogue scene
S1→S2
3D CGI · 2-Shot · Complex Action
Cut-In: Fighting sequence with strike
S1→S2
Live Action · 4-Shot · Multi-Character
Group Shot → Two Shot → Close-Up
S1→S2
Anime · 2-Shot · Simple Action
Over-Shoulder: Character handing object
S1→S2
Live Action · 5-Shot · Complex Action
Dance choreography, Low-Angle series
S1→S2
Documentary · 2-Shot · Simple Action
Cut-Out: Subject exiting frame, Wide Shot
S1→S2
3D CGI · 3-Shot · Medium Action
Walk cycle through environment, pan-follow
ACT Ⅵ
BibTeX
@article{zhuang2026act2cut,
title={Act2Cut: Continuous Next-Shot Video Narrative Match on Action-Cut},
author={Zhuang, Cailin and Hu, Yaoqi and Dong, Zheng and Zhang, Shiwen and Huang, Haibin and Zhang, Chi and Li, Xuelong},
journal={ACM Transactions on Graphics (TOG)},
volume={1},
number={1},
year={2026},
publisher={ACM New York, NY, USA}
}