PM-VIS+: High-Performance Video Instance Segmentation without Video Annotation
Zhangjing Yang, Dun Liu, Xin Wang, Zhe Li, Barathwaj Anandan, Yi Wu
TL;DR
PM-VIS+ tackles video instance segmentation without manual video annotations by training on image datasets and supplementing missing categories with ImageNet-bbox. It introduces a three-stage pipeline: generate pseudo-labeled video data from an image-trained PM-VIS+(Image), refine those labels with DeAOT and score-based filtering, and train PM-VIS+(Video) on the optimized data. The approach uses dynamic supervision signals that combine pixel-level and bounding-box annotations, enabling VIS performance competitive with fully supervised methods while reducing labeling costs. Experiments across COCO, YTVIS, and OVIS demonstrate effective gains from pseudo-label optimization and backbone scaling (ResNet-50 and Swin-L), highlighting practical impact for data-constrained VIS deployment.
Abstract
Video instance segmentation requires detecting, segmenting, and tracking objects in videos, typically relying on costly video annotations. This paper introduces a method that eliminates video annotations by utilizing image datasets. The PM-VIS algorithm is adapted to handle both bounding box and instance-level pixel annotations dynamically. We introduce ImageNet-bbox to supplement missing categories in video datasets and propose the PM-VIS+ algorithm to adjust supervision based on annotation types. To enhance accuracy, we use pseudo masks and semi-supervised optimization techniques on unannotated video data. This method achieves high video instance segmentation performance without manual video annotations, offering a cost-effective solution and new perspectives for video instance segmentation applications. The code will be available in https://github.com/ldknight/PM-VIS-plus