two-stream-action-recognition icon indicating copy to clipboard operation
two-stream-action-recognition copied to clipboard

spatial_cnn overfitting

Open Robinxin123 opened this issue 5 years ago • 5 comments

hello all, when i run the spatial_cnn.py, i found the loss function for test is very strange. does someone have the same problem?
many thanks!

overfitting

Robinxin123 avatar May 17 '19 09:05 Robinxin123

Hello, I am preparing to run this model. I don't know what this specific environmental requirement is. I want to ask you, thank you very much!

Songyyan avatar May 24 '19 03:05 Songyyan

我也碰到了过拟合的问题,请问楼主后来解决了吗?

xqqxiaoqiqi avatar Apr 12 '20 04:04 xqqxiaoqiqi

@Robinxin123 Can you tell me the learning rate you are keeping?

I am getting these result, prec@1 and prec@2 on just two epochs but this is really so strange?

Capture4

Tariq195 avatar Aug 17 '21 05:08 Tariq195

Should I run all 500 epochs for training only? then check the validation or should I run both? because the epoch for validation is taking 21-22min and 19-20 mins for training, add them it will be 40mins, for each epoch it is taking 40 minutes. If I run it for 500 epochs it will take almost 13 days for training validation, I am using RTX 3080 10 GB + i9 10th generation.

Tariq195 avatar Aug 17 '21 06:08 Tariq195

您好,我遇到的问题是FileNotFoundError: [Errno 2] No such file or directory: 'dic/frame_count.pickle',我该如何得到这个.pickle文件

202122209010 avatar Mar 21 '23 12:03 202122209010