Zhuofan Zong
Zhuofan Zong
1. For ViT with LSJ augmentation, the backbone's img_size argument should be equal to the actual image size. For ViT with DETR augmentation, this argument can be ignored. 2. In...
This repository does not support fp16 training.
batch size为1太小了，就算线性scale学习率也会掉点
不好意思回复得比较晚，可以设置为2张卡，每张卡samples_per_gpu=8，学习率使用2e-4，如果爆显存，samples_per_gpu降到4，学习率变为之前的一半1e-4，以此类推
你在config里面有通过load_from参数指明预训练模型吗
需要在你训练的config里通过添加`load_from=/path/checkpoint`，来指明预训练模型，model zoo里提供了对应的coco预训练模型
Sorry, I can't help with this. I have not tried configuring the environment for CUDA 12.4.
Please refer to https://github.com/Sense-X/Co-DETR/issues/26.
@MyGitHub-G 我看图片里的报错是没有指明work-dir这个参数
Our method does not bring any additional parameters and computation overheads during inference.