【CV论文】SOLO算法

前言

论文

SOLO: Segmenting Objects by Locations,
Xinlong Wang, Tao Kong, Chunhua Shen, Yuning Jiang, Lei Li
In: Proc. European Conference on Computer Vision (ECCV), 2020
arXiv preprint (arXiv 1912.04488)

SOLOv2: Dynamic and Fast Instance Segmentation,
Xinlong Wang, Rufeng Zhang, Tao Kong, Lei Li, Chunhua Shen
In: Proc. Advances in Neural Information Processing Systems (NeurIPS), 2020
arXiv preprint (arXiv 2003.10152)

参考的代码为在mmdetection的基础上实现的:https://github.com/WXinlong/SOLO

 

ResNet + FPN + solov2Head + maskFeatHead;
训练的时候,金字塔特征之后类别过滤一次,特征过滤一次,还有NMS过滤一次,得到结果之后计算loss,分别计算focalLoss和DiceLoss;
测试的时候,金字塔特征之后类别过滤一次,特征过滤一次,还有NMS过滤一次,之后进行上采样插值,concate之后输出结果;
train: FPN -> conv -> nms —> loss;
infer: FPN -> conv -> nms -> filter(两部分) -> 上采样插值interpolation -> concate (模型输出) -> 模型输出结果进行filter;

参考

1. 实例分割新思路之SOLO v1&v2深度解析

2. SOLO: Segmenting Objects by Locations

3. SOLOv2: Dynamic and Fast Instance Segmentation

4. github_SOLO;

 

 

posted on 2021-10-31 21:40  鹅要长大  阅读(191)  评论(0编辑  收藏  举报

导航