Jas
Jas
Please set `flip_test=False` in the configs for higher speed.
Thanks for reporting @fabro66. Will try this out.
我们之后也想一下这个case,怎么处理更好一点:)
Thanks a lot for your contribution! We will review the PR ASAP. @walsvid Please sign the CLA so that we can proceed the code merging.
你觉得哪里有异常呀? 训练久一点会不会有值呢?
哦 L,M两个都是-1 有点奇怪了,Large 和 Medium的物体都是不存在的?
training acc 有值么? 如果有的话,可以看看可视化效果正不正常
想起一个可能的原因,你可以检查下标注文件的格式,有没有漏写这个categories? https://github.com/jin-s13/xtcocoapi/blob/e1052f3009702b74ece8aa58e3ef8b73d10467f2/annotations/example_coco_val.json#L49
The simplest way to achieve this is to build a single big head, containing multiple sub-heads. Modify the forward function of the single big head, to inference these sub-heads.
Thanks @ailingzengzzz , is this PR ready for review now?