关于数据预处理，求大佬解答
yinwang950613 opened this issue · comments
大佬，能问下这个ilsvrc_vid.txt and youtube_final.txt是怎么生成的吗？
在dataset里面有ipynb
那为什么在训练的时候，就是运行traning代码的时候，还需要download checkpoint.pth.tar?
那个是load pretrained vggnet 的参数 做transfer learning
这样子啊，谢谢你！
还有一个问题，既然是transfer learning ,那代码中的 mod = models.vgg16(pretrained = True)这个不就是把VGG的网络参数给自己定义的新的model吗？这两者有什么不一样呢？
等等。。。download checkpoint.pth.tar在哪一行？
我以为你说的是mod = models.vgg16(pretrained = True)
就是在运行traning代码的时候，指令后面有个arg的参数 ，叫做 --pre 对吧，python train.py 0 SIAMVGG --pre='backup/bestSIAMVGG.pth.tar' ，这个--pre是指所训练model的保存路径？还是说下载的vgg16的.pth，不太明白，因为我运行这个代码时会出现错误，IOError: [Errno 2] No such file or directory: 'backup/siamVGG_ywcheckpoint.pth.tar'
Epoch: [0][119/125] Time 0.278 (0.292) Data 0.247 (0.258) Loss 263.2892 (490.3870)
Epoch: [0][120/125] Time 0.312 (0.292) Data 0.282 (0.259) Loss 270.5379 (488.5701)
Epoch: [0][121/125] Time 0.286 (0.292) Data 0.255 (0.259) Loss 274.4255 (486.8148)
Epoch: [0][122/125] Time 0.291 (0.292) Data 0.260 (0.259) Loss 226.7929 (484.7008)
Epoch: [0][123/125] Time 0.290 (0.292) Data 0.258 (0.259) Loss 310.4707 (483.2957)
Traceback (most recent call last):
File "/home/tanjianhao/yinwang/procedure/SiamVGG-master/train/train.py", line 259, in
main()
File "/home/tanjianhao/yinwang/procedure/SiamVGG-master/train/train.py", line 136, in main
}, is_best,args.task)
File "/home/tanjianhao/yinwang/procedure/SiamVGG-master/train/utils.py", line 62, in save_checkpoint
torch.save(state, 'backup/'+task_id+filename)
File "/home/tanjianhao/anaconda3/envs/pytorch_py2.7/lib/python2.7/site-packages/torch/serialization.py", line 224, in save
Epoch: [0][124/125] Time 0.298 (0.292) Data 0.267 (0.259) Loss 270.0476 (481.5897)
return _with_file_like(f, "wb", lambda f: _save(obj, f, pickle_module, pickle_protocol))
- best MAE 0.000
File "/home/tanjianhao/anaconda3/envs/pytorch_py2.7/lib/python2.7/site-packages/torch/serialization.py", line 147, in _with_file_like - MAE 0.000
f = open(f, mode)
IOError: [Errno 2] No such file or directory: 'backup/siamVGG_ywcheckpoint.pth.tar'
当运行traning之后，开始会出现训练的迭代的信息，但是在epoch[0]迭代到123的时候，就会报错，no such file or dirrectory: ...
你要新建一个目录叫backup.
@leeyeehoo 为什么我直接用你的vgg网络去替换alexnet进行训练得到的loss是nan呢（siamfc代码用的是别人的）
请问VID是2015还是2017
@yinwang950613 what the first number and the end number of you loss in training siamVGG?
大佬，能问下这个ilsvrc_vid.txt and youtube_final.txt您有吗，我对数据预处理这块不太懂，能够分享一下您的.txt文件。