xiyufeng2

Results 15 comments of xiyufeng2

I have seen that already,but not helpful.I have biult Detectron2 as instruction。

only contains 4 files: Imagesets、 examples、README、data_prepare.py。no trained models

found it,thank you! the providied file "r-yolov3-polygon.pt" Should Pt be PTH?

there is no trained models in your connection,where can I find it?

感谢您的回复! CUDA_LAUNCH_BLOCKING=1调试后报错位置为rand_inds = gallery[perm],在random_choice函数中。首先这几个张量都无法打印内容,看不到里面是什么:其次查了网络结构,最后的分类fc层确实是16,DOTA数据集是15类,难道是这里对应不上吗

已解决,随机生成perm时超出范围了。感谢

I didn't load pretrained weight to train the model , leading the loss to be large. I think the weight init method is not suitable

我也有相同的疑问。训练自己的数据,修改了配置文件dota.yaml,自己的数据也用txt2json转换的格式。训练时optim.step()报错,:维度不匹配:16与2(我自己的一类)。查看了网络输出,最后是2分类没问题,是否需要在paths_catalog.py中自己建一个类似于coco的类???这也太麻烦了。我不能自己在dota的工厂上改吗

最后梯度反传的时候告我类别不匹配!这单步调试还找不出来毛病

已及解决:多式几次就好了