Colin Xiang
Colin Xiang
Hi, 我测试了TAT数据集中的Family,使用的是默认脚本test_tnt.sh,采用normal融合,最近仅得到13MB点云文件。经检查发现生成的mask文件夹中的_geo.png都是大部分区域黑色图片,从而最后得到的 final.png的大部分区域都是无效的。geometric consistency阈值分别是默认的0.01和1。不知道您这边是否有一样的问题?
f-score
Hi, 我发现f-score计算,DTU没有提供,T&T提供了在线提交接口。你这边可以分享下DTU的f-score计算方法吗?
您好,发现TranMVSNet论文中描述,在DTU和BlendedMVS分布训练时候,focal loss 中的γ参数分别为0(适用于simple scenarios)和2(适用于complicated scenario). 但是对照代码发现trans_mvsnet_loss_fl 和focal_loss_bld函数中的γ都是0 ?
Hi, I follow the instruction in ReadMe.md to train "Brandenburg Gate (scale1)" on two A100 GPUs, but the mesh result is bad than your paper. It is very strange that...
Hi, I have noticed that camera model exported from "colmap_refined" folder is ”SIMPLE_RADIAL“. May I ask whether code support different camera model (such as SIMPLE_PINHOLE, PINHOLE) ? 
您好,方便问下TAT数据集里面inter和advanced的每个场景的阈值参数您这边是如何调整的呢?(主要是光度一致性和几何一致性参数)
Hi, I found the the structure of raw building scene from Mill-19 has following directory, which is different from tanks and temples. building (Mill-19) --train --metadata --rgbs --val --metadata --rgbs...