PMRS-lab
PMRS-lab
我在对比官方发布的superpoint网络代码时,发现你的代码中训练部分当不加bn层时,效果会变的很差,而官方发布的网络中并没有在每个卷积模块后带bn层。我在实测你的网络superpoint的描述子时发现你的描述子的描述能力与superpoint官网并且和其他人复现的代码的描述子的描述能力差距很多,且同名点描述效果差距很大,请问你自己有实测吗?
好的,谢谢,但是从目前实测效果来看你版本的superpoint训练代码与官方发布效果以及我在github上测试的其他版本的测试效果还是差挺多的
> Hello, where is the download link for the tourism_data dataset? 您好,请问您有找到tourism_data 数据集的下载链接地址吗?我也遇到了同样的问题
> I would like to utilize the DinoV2 model pretrained on the ImageNet-1k dataset as the backbone for my research, specifically for comparing it with other methods in my research...
> > > If you use a single node, it will automatically default to DDP training. > > > > > > if i have a PC which consisting of...
@patricklabatut @vkhalidov @niavasha @leo-gan @goggle Excuse me, May I ask how to use the file dinov2_vitl14_pretrain.pth and fine-tune it again based on my dataset? There is only backbone information in...