RSNA数据集
0xTechSavvy opened this issue · 2 comments
0xTechSavvy commented
我用RSNA的数据集进行按照默认参数训练,阈值是 [0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75]。
结果detnet的效果要好多detnet-cascade
detnet在6epoch,验证集map达到最好:
mAPs: [0.7224553391260808, 0.6599297035067196, 0.579511199465727, 0.48372004238688515, 0.377126303947345, 0.27479695269073545, 0.19356904730168234, 0.13584048094266832]
mean mAP: 0.4283686336709805
detnet-cascade在5epoch,验证集map达到最好:
mAPs: [0.6793144470700885, 0.5995099883468915, 0.5322867614959018, 0.44291050802950327, 0.3290775815062264, 0.257612591509458, 0.1926359629090184, 0.14070865806833324]
mean mAP: 0.39675706236692765
@guoruoqian 这是什么原因呢?
gittigxuy commented
@hongdayu ,请问如何训练自己的数据集?你将RSNA数据集转化为coco格式还是voc格式了呢?
0xTechSavvy commented
@hongdayu ,请问如何训练自己的数据集?你将RSNA数据集转化为coco格式还是voc格式了呢?
voc啊