Xinwei Dai
Xinwei Dai
> you can follow the repo https://zhuanlan.zhihu.com/p/147144376, download the "bert-base-uncased" folder from https://huggingface.co/bert-base-uncased, and modify the error pretrain location to your local position. I downloaded it locally this way, Will...
Hello author, I use the Segment Anything of SAMed_h, but I load the weight of vit-b, is there any difference between this and the weight of vit-b loaded on SAMed...
Since I am a low-level task, if the resize image will damage the image information, whether the resize position embedding will affect the performance? I have another question. Is promptLayer...
> 我最近在学习大大们的优秀的工作,在图像输入时做对齐时有一些问题想要请教: 1.为何图像在做焦距对齐时,做焦距倍率缩放时主点(cx,cy)没有做变化。在正常投影关系里,如果cx,cy没有变化对应的相机坐标系下的图像应该是裁剪后的结果吧,这个也比较符合直觉。举例的话,如焦距ratio为2时,把(100,100)的图像resize到(200,200),同时crop中心的(100,100)对应的才是cx,cy不变同时fx,fy*2的内参的投影结果。 2.为了把对齐内参的图像输入到网络,还需要进行一次scale的对齐。同样的,为什么这次缩放又仅仅进行了主点的缩放呢?每次对图像的resize应该都会对对应相机模型的内参产生同样的变化吧,为何这次又只进行了cx,cy的缩放呢? 新手学习,可能提到的问题不甚全面,希望大大可以帮忙答疑解惑,万分感谢! 我不知道理解的对不对,depth 已经按 1/resize_ratio 缩放(metric depth 修正) 图像像素被 resize → 对应的 cx, cy 要跟着移动,保证投影像素正确 fx, fy 表示相机焦距 单位为像素 由于 depth 已经缩放成 metric depth / resize_ratio fx,...
"I encountered the same issue, and the inference result is !!!!!!!!!!!!."
> can anyone help me on this? Have you solved this problem?
> > > 有人能帮我吗? > > > > > > 你解决了这个问题吗? > > 我也这样 我已经解决这个问题了
I have the same need. Has anyone solved it?
> Hello 你好 > > Thank you for your great work感谢您的出色工作 > > I want to finetune on my own data. I have some question我想对自己的数据进行微调。我有一些问题 > > 1. I found...
> Why is Moge's detail geometry so good? Does its detail geometry originate from the Affine-invariant point map or the Multi-scale local geometry loss? For example, in an autonomous driving...