闫令

How about the linear evaluation on ImageNet

3

How about the linear evaluation on ImageNet? ^_^

There are some bug when load the optimizer and scheduler in checkpoint, we change the "train_net.py" according to the codes in "FreeSOLO" that committed by Xinlong Wang.

the model will converge faster than RAFT?

2

Since the model is initialized by the coarse flow instead of zero flow, will it converge faster than RAFT? Maybe it will get good performance with fewer training steps? thank...

能否集成到vscode的插件里

10

如题^_^

请问，RGB 和BGR以及mean，std是怎么对应的

in paper,you concat images and segmentaion maps as input pf discriminstor, but in your codes ,i found not the images concat with seg maps. Where can i found it ?...

求一下2.0的数据集

3

想要一下2.0的数据集，麻烦可以分享一下吗，感谢[email protected]

随机不确定

3

有没有人在训练时遇到每次训练结果不一样的情况，在一开始已经设定了随机种子固定，调试发现，模型在第一次更新参数前是完全一致的，相同的输出产生相同的预测、loss，但是经过一次参数更新后，模型参数和预测就不一样了，这是哪里还有随机的部分没被固定？ def set_seed(random_seed): random.seed(random_seed) # NumPy np.random.seed(random_seed) # PyTorch torch.manual_seed(random_seed) # if torch.cuda.is_available(): torch.cuda.manual_seed(random_seed) torch.cuda.manual_seed_all(random_seed) torch.backends.cudnn.deterministic = True torch.backends.cudnn.benchmark = False

SCD 2023论文

1

请问SCD2023这篇论文中，公式8的x_i,j,k,l指的是什么，为什么会有四个下标呢？我在论文中好像没有看到具体的说明