arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2205.02101cs.CVcs.AIcs.LG
收藏

动态稀疏R-CNN

Dynamic Sparse R-CNN

Advanced Micro Devices, Inc.(超威半导体公司)

Qinghang Hong, Fengming Liu, Dong Li, Ji Liu, Lu Tian, Yi Shan

摘要:

稀疏R-CNN是最近通过对稀疏、可学习的建议框和建议特征进行集合预测的强大目标检测基线。在这项工作中,我们提出用两种动态设计来改进稀疏R-CNN。首先,稀疏R-CNN采用一对一的标签分配方案,其中匈牙利算法仅用于为每个地面真值匹配一个正样本。这样的一对一分配对于所学习的建议框和地面实况之间的匹配可能不是最佳的。为了解决这个问题,我们提出了基于最优传输算法的动态标签分配(DLA),以在稀疏R-CNN的迭代训练阶段分配越来越多的正样本。我们限制匹配逐渐松散的顺序阶段,因为后期阶段产生的改进建议,提高精度。其次,在稀疏R-CNN的推理过程中,学习到的建议框和特征对于不同的图像保持固定。受动态卷积的启发,我们提出了动态建议生成(DPG),以动态地组装多个建议专家,为连续的训练阶段提供更好的初始建议框和功能。因此,DPG可以导出样本相关的建议框和特征以进行推理。实验表明,我们的方法,名为动态稀疏R-CNN,可以提高强大的稀疏R-CNN基线与不同的骨干目标检测。特别是,Dynamic Sparse R-CNN在COCO 2017验证集上达到了最先进的47.2% AP,在相同的ResNet-50主干上超过了Sparse R-CNN 2.2% AP。

英文摘要:

Sparse R-CNN is a recent strong object detection baseline by set prediction on sparse, learnable proposal boxes and proposal features. In this work, we propose to improve Sparse R-CNN with two dynamic designs. First, Sparse R-CNN adopts a one-to-one label assignment scheme, where the Hungarian algorithm is applied to match only one positive sample for each ground truth. Such one-to-one assignment may not be optimal for the matching between the learned proposal boxes and ground truths. To address this problem, we propose dynamic label assignment (DLA) based on the optimal transport algorithm to assign increasing positive samples in the iterative training stages of Sparse R-CNN. We constrain the matching to be gradually looser in the sequential stages as the later stage produces the refined proposals with improved precision. Second, the learned proposal boxes and features remain fixed for different images in the inference process of Sparse R-CNN. Motivated by dynamic convolution, we propose dynamic proposal generation (DPG) to assemble multiple proposal experts dynamically for providing better initial proposal boxes and features for the consecutive training stages. DPG thereby can derive sample-dependent proposal boxes and features for inference. Experiments demonstrate that our method, named Dynamic Sparse R-CNN, can boost the strong Sparse R-CNN baseline with different backbones for object detection. Particularly, Dynamic Sparse R-CNN reaches the state-of-the-art 47.2% AP on the COCO 2017 validation set, surpassing Sparse R-CNN by 2.2% AP with the same ResNet-50 backbone.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑