huichen98

Results 9 comments of huichen98

train own dataset, but Found 0 boxes for img

The problem is your voc_annotation.py file. In this file，you should modify your classes which belong to you！

bevformerv2 二阶段

是的

KeyError: '1b9a789e08bb4b7b89eacb2176c70840'

没有的直接跳过

去噪任务的影响

感谢答复，我们还发现在角度预测上 sparsebev 要优于bevformer，bevformer采样点虽然很密集，但由于是固定点，似乎收益不大

有关perspective_transform和pixels_per_meter这两个参数问题

需要标定获取相机的内外参数

Can't generate results on single gpu test

> > I recommend you to use this script './tools/dist_test.sh ./projects/configs/bevformer/bevformer_base.py ./path/to/ckpts.pth 1' > > I have tried this command early time. But the problem still remains. You should first...

CUDA error: device-side assert triggered

我也一样报错

training error

我取出数据集前1000帧，打开或者关闭增强，训练都没有报错，但是用作者提供的整个数据集训练，就会出现错误

training error

config --> fp16 = dict(loss_scale="dynamic")