full data sup only 设置下batchsize=2*8,iter=12epoch~=88k; semi设置下batchsize=sup 4*8+unsup 4*8 ,iter=240k;

JCZ404 / Semi-DETR

[CVPR 2023] Official implementation of the paper "Semi-DETR: Semi-Supervised Object Detection with Detection Transformers"

https://arxiv.org/abs/2307.08095

MIT License

77 stars 9 forks source link

Closed lixiangamy closed 5 months ago

lixiangamy commented 10 months ago

我不确定理解是否有问题？如果没问题，semi设置下labeled data训练iter比sup only多很多？

JCZ404 commented 5 months ago

是的，确实是这样，因为labeled data和unlabeled data两者数据量不同。