Publications
Publications and preprints by Wang Chen on long-video understanding, multimodal reasoning, and generative vision.
2026CVPR 2026
Detects semantic boundaries from a noisy query-frame relevance signal, then allocates frame budgets by segment importance. The training-free method improves VideoMME, MLVU, and LongVideoBench by 5.5, 9.5, and 6.2 points.
2026AAAI 2026 · Equal contribution
Allocates visual tokens to frames before cross-modal interaction. CoT turns a complex query into scorable clues, improving six video benchmarks by 3.2 points on average under the same token budget.
2023
2026ICML 2026
Combines off-the-shelf modality experts without additional training through an LLM controller, textualized cross-modal memory, and a full-duplex interaction layer.
2026arXiv preprint · Equal contribution
Partitions video into coherent events, localizes a query-relevant anchor in each event, and applies adaptive MMR for global refinement across coverage, relevance, and diversity.
2024SSRN preprint
Constructs an aesthetics hyperplane and interpolates directly in latent space for real-time, target-score-controlled editing with identity preservation.
2026arXiv preprint
Evaluates audio-visual social interaction through speaker perception, interruption timing, and response behavior, exposing gaps between perception accuracy and interaction quality.
2026arXiv preprint
Uses spatial-temporal decoupled wavelet analysis to allocate frame-level budgets and compress spatial tokens while preserving performance at high compression ratios.
2026arXiv preprint
Produces a single frame priority ranking that can be truncated at any budget, progressively expanding from local evidence to temporal context while avoiding repeated selection.
Author order and publication status are checked against the paper pages. * denotes equal contribution. Preprints are not labeled with withdrawn or pending submission venues.