Our study presents a comprehensive comparison of our Uni-OVSeg against previous works across a range of benchmarks, showing significant performance improvements in open-vocabulary semantic and panoptic segmentation tasks.
In the PASCAL Context-459 dataset, Uni-OVSeg not only surpasses weakly-supervised counterparts but also outperforms the cutting-edge fully-supervised methods, indicating its superior capability in categorizing diverse semantic classes.
Uni-OVSeg achieved enhancements of 18.3% and 12.2% mIoU in PASCAL VOC with 20 and 21 classes, respectively, demonstrating its ability to capture fine-grained spatial structures.
The results presented elevate the practical applicability of weakly-supervised open-vocabulary segmentation methods, showcasing the transformative potential of our Uni-OVSeg framework in the field.
#open-vocabulary-segmentation #semantic-segmentation #panoptic-segmentation #weakly-supervised-methods #deep-learning-advancements
Collection
[
|
...
]