AlphaPlusTT

Not yet, but i can give a try :)

any suggestion for the IOError？

在真实数据上测试模型，且都在imagenet上进行模型权重初始化，真实数据和虚拟数据混在一起训练和拿虚拟数据pretrain然后在真实数据上finetune相比，哪种方式更好？作者有试过类似的尝试吗？

i meet the same error when i use the pytorch1.0 and the error disappear when i use the pytorch1.1

@yangz0626 i found one inplace operation in ssd_rotate_head.py:327, maybe you can modify this line and have a try

Could it be the difference between softmax and sigmoid that affects the results？

@weiyithu @Xpangz Hi there, why is the distance between neighboring voxels in the ground truth set to 1? Considering that the voxel_size is 0.5 during both the training and inference...

1： 发现作者的回答并不是真正的原因，虽然`groundtruth`和`prediction`的尺度不一致，比如说前者的一个维度是0-199步长是1，后者的一个维度是-49.75-49.75步长是0.5，但是在visual.py中的这个地方已经通过`* voxel_size`调整了步长： https://github.com/weiyithu/SurroundOcc/blob/f698d7968a60815067601776dabfca2a8b03500a/tools/visual.py#L43 并且在这里确定了调整步长后的尺度： https://github.com/weiyithu/SurroundOcc/blob/f698d7968a60815067601776dabfca2a8b03500a/tools/visual.py#L58 因此修复bug以后这里的`voxel_size`可以设置为任意值： https://github.com/weiyithu/SurroundOcc/blob/f698d7968a60815067601776dabfca2a8b03500a/tools/visual.py#L36 2： 这里之所以可视化的时候会出现“稀疏”的现象是因为groundtruth的数据类型是'int64'，而这里取数组的一部分进行浮点运算无法改变原数组的数据类型： https://github.com/weiyithu/SurroundOcc/blob/f698d7968a60815067601776dabfca2a8b03500a/tools/visual.py#L43 举个例子: `int((1+0.5)*0.5)=int(0.75)=0` 然而作者的意图是想得到0.75，并不是得到0，所以才导致了可视化时“稀疏”的现象。 3： 修复的方法很简单，在下面的代码前面加上`fov_voxels = fov_voxels.astype('float')`即可： https://github.com/weiyithu/SurroundOcc/blob/f698d7968a60815067601776dabfca2a8b03500a/tools/visual.py#L43

@weiyithu Thank you for your detailed explanation. My dataset includes four cameras, but it's not a complete 360° setup. Currently, I've only calibrated the front camera and would like to...

Thank you for your patient answer. Would you mind sharing more information about the angles between cameras and the installation positions of cameras in the in-the-wild data mentioned in README?...