X2SAM is a unified multimodal large model that combines image and video segmentation with language and visual prompts, introduces a Mask Memory for temporal consistency, defines a new V‑VGD task, and achieves state‑of‑the‑art results while cutting training cost by over 30%.
DVIS introduces a decoupled video instance segmentation framework that splits the task into segmentation, tracking, and refinement modules, achieving state-of-the-art performance across VIS, VPS, and VSS benchmarks while maintaining low computational overhead, and demonstrates robustness in both online and offline settings.
The paper presents Video Object of Interest Segmentation (VOIS), a new e‑commerce task that locates and segments video instances matching a given product image, introduces the LiveVideos dataset of 2,418 Taobao live‑stream clips, and proposes a dual‑path Swin‑Transformer with cross‑fusion modules that outperforms existing VOS/VIS baselines.
Meituan’s six CVPR 2022 papers advance computer vision by introducing a few‑sample model compression method, a language‑bridged video object segmentation approach, a single‑stage 3D visual grounding technique, a dynamic early‑exit image captioning system, a boosted black‑box adversarial attack, and a semi‑supervised video paragraph grounding framework.
Kuaishou secured 14 accepted papers at CVPR 2021, spanning 3D hand mesh recovery, unsupervised keypoint detection, point cloud completion, modular interactive video segmentation, deep video matting, co‑salient object detection, occlusion‑aware instance segmentation, semantic image matting, and face forgery detection, showcasing the maturity of its research collaborations.
MiVOS introduces a highly decoupled, three‑module framework—Interaction‑to‑Mask, Mask Propagation, and Difference‑aware Fusion—for interactive video object segmentation, delivering precise masks with fewer user interactions, validated on the DAVIS benchmark and supported by a new large‑scale synthetic VOS dataset (BL30K).
This technical talk introduces video object segmentation tasks, reviews leading datasets and state-of-the-art deep learning models, and shares practical network design rules and performance‑boosting techniques, presented by Prof. Wang Xinggang as part of Alibaba's MEDIA AI challenge series.
The MEDIA AI Alibaba Entertainment Algorithm Challenge, co‑hosted by Alibaba Entertainment Technology and Alibaba Cloud Tianchi, invites researchers and teams to tackle high‑precision video person segmentation and video temporal event detection using large‑scale datasets, with a total prize pool of 380,000 CNY and recruitment opportunities.
The paper proposes a 3D‑CNN scoring system that ranks short, information‑dense video segments, selects the most exciting one, and converts it into a looping GIF cover, replacing static thumbnails; trained on large video‑GIF datasets with pairwise ranking loss, it improves click‑through rates while reducing bad‑case generation.
Ren Haibing’s Youku Cognitive Lab talk reviews object segmentation’s motivation, explains semantic and instance concepts, presents UNet‑based and category‑agnostic methods—including fast video segmentation with motion cues—and reports high IoU results while outlining future edge‑aware, label‑free, and non‑online video segmentation research directions.
