Google Object Detection SSD_inception_V2架构

时间:2018-10-02 16:13:38

标签: tensorflow object-detection-api

[问题] 论文:现代卷积目标检测器的速度/精度权衡

该论文说:“我们使用混合4c和混合5c,在四个附加的卷积层上附加 分别降低深度512、256、256、128的分辨率。”但是,我不完全了解架构。这意味着混合4c->混合5c-> 512-> 256-> 256-> 128的顺序?我也引用了代码,但仍然没有。只是,我知道SSD_inception_V2功能提取器结构('Conv2d_1a_7x7','MaxPool_2a_3x3','Conv2d_2b_1x1','Conv2d_2c_3x3','MaxPool_3a_3x3','Mixed_3 Mixed_4a','Mixed_4b','Mixed_4c','Mixed_4d','Mixed_4e','Mixed_5a','Mixed_5b','Mixed_5c')。

https://github.com/tensorflow/models/blob/master/research/slim/nets/inception_v2.py

[参考代码站点] 关键字:from_layer

https://github.com/tensorflow/models/blob/1f484095c0981e2a62403b16256cb877749dfe94/research/object_detection/models/feature_map_generators.py https://github.com/tensorflow/models/blob/master/research/object_detection/models/ssd_inception_v2_feature_extractor.py

0 个答案:

没有答案