Gangwei XU

可以多下载几次。偶尔会卡住2024-01-26 15:49:25>hukangli 写道： 0.9.7的时候我也报错，0.5.4就不报了。中间的没试过 @BayMaxAC 你好，我在降低了timm版本为0.5.4后，运行训练脚本没报错，但是一直卡在模型下载这一步。请问您能提供下载完成的mobilenetv2_100_ra-b33bc2c4.pth模型吗？ — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID:...

OK，good luck！ 发件人："hukangli" ***@***.***> 发送日期：2024-01-26 16:45:12 收件人："gangweiX/IGEV" ***@***.***> 抄送人： 主 题：Re: [gangweiX/IGEV] timm库加载模型时报错 (Issue #43) 多下载几次成功了 — Reply to this email directly, view it on GitHub, or unsubscribe. You are...

It is trained on sceneflow with data augmentation.

Thank you for your attention, this is caused by the inconsistency of the timm version. Please pip install timm==0.5.4. I have introduced how to install environment in README.md.

是的，在论文中，我们只测试了泛化性，用的是在Scene Flow上预训练的模型2023-12-18 13:28:56>Wade_Lucky 写道： 作者您好，我想咨询您一下，在这两个数据集上，您论文中的指标直接是用的Sceneflow中的预训练模型进行检测的吗？而不是混合微调以后的结果！感谢您的指导 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

如果你把k设成att_weights的dim=2的维度值（也就是48），那确实没必要两个sort和gather，这里的sort和gather主要是为了构建一个topk的attention volume，构建一个稀疏的代价体，降低后面代价聚合的计算开销

Thanks, you may refer to https://github.com/antabangun/coex, In this repo, you can find demo.py.2023-11-06 10:03:32>"小艾同学" 写道： Excellent work! But I just wonder how can I use this network to process video...