Open dupc2018 opened 2 years ago
确保你训练测试时embt都是对应正确的
训练测试的embt都是你原来的都没动,也就是数据集换了一下,其他的都没变,结果输出的7帧全部都是一样的,而且结果不怎么清晰,不知道原因出现在什么地方了
建议先使用train_gopro.py
在GoPro数据集上训练并测试,不要用train_vimeo90k.py
训练的模型在demo_8x.py
脚本中进行测试。
直接使用train_gopro.py加载GoPro数据集是不是有点问题呢,我使用自己的数据集用train_gopro.py训练的,测试是用demo_8x.py脚本测试的,造成输出的都一样,连输出的图片文件大小都一样
是不是demo_8x.py的原因呢,写上False 加载模型的时候 model.load_state_dict(torch.load('/mnt/IFRNet-main/checkpoint/IFRNet_L/2022-06-26 02:14:46/IFRNet_L_latest.pth'),False),但是如果不写False就报错 RuntimeError: Error(s) in loading state_dict for Model: Missing key(s) in state_dict: "encoder.pyramid1.0.0.weight", "encoder.pyramid1.0.0.bias", "encoder.pyramid1.0.1.weight", "encoder.pyramid1.1.0.weight", "encoder.pyramid1.1.0.bias", "encoder.pyramid1.1.1.weight", "encoder.pyramid2.0.0.weight", "encoder.pyramid2.0.0.bias", "encoder.pyramid2.0.1.weight", "encoder.pyramid2.1.0.weight", "encoder.pyramid2.1.0.bias", "encoder.pyramid2.1.1.weight", "encoder.pyramid3.0.0.weight", "encoder.pyramid3.0.0.bias", "encoder.pyramid3.0.1.weight", "encoder.pyramid3.1.0.weight", "encoder.pyramid3.1.0.bias", "encoder.pyramid3.1.1.weight", "encoder.pyramid4.0.0.weight", "encoder.pyramid4.0.0.bias", "encoder.pyramid4.0.1.weight", "encoder.pyramid4.1.0.weight"
You can search this error and get the solution, such as this issue discussion. Also, I have fixed this problem by changing ddp_model.state_dict()
to ddp_model.module.state_dict()
when saving model checkpoints.
For the problem that all output frames are the same, I suggest that you can read the paper and run provided code to understand the working principle of IFRNet, then you should know how to solve it.
我也有这个问题。。啥情况?
测试的时候出现: 运行demo_8x.py 结果 out1 到out7 一致的问题,是训练的问题还是哪里出现问题了呢