search

pochih / Video-Cap

🎬 Video Captioning: ICCV '15 paper implementation
48 stars 20 forks source link

why are all the test results the same? #3

Open XiangChen1994 opened 6 years ago

XiangChen1994 commented 6 years ago

I trained the network for 3320 epochs, but when I test the model the results are mostly the same, which show like below. .... video => rl1rVk_xIOs_1_16.avi generated_sentence => a man is playing a flute reference => Children are dancing. .... video => dfOuTx66bJU_34_39.avi generated_sentence => a man is playing a flute reference => A man walking down the street turns around, then turns back and starts jogging. .... video => f9Won2JpOEU_60_80.avi generated_sentence => a man is playing a flute reference => A cat is laying down licking it's paw. .... what is the reason for this situation?

pochih commented 6 years ago

There may be some bugs, let me check it.

nayyeraafaq commented 6 years ago

Hi, are you able to fix the bug ?? I am getting the same error. All the outputs are almost same and looks like network is not learning anything. The outputs are like "a a a is", "a a is is a", "a a is" etc. What's wrong with the model ??

XiangChen1994 commented 6 years ago

I got the error because there are some problems in my OpenCV which caused the extracted features are wrong. It was solved after I changed my OpenCV.

srilekha1993 commented 4 years ago

@nayyeraafaq i am also getting same captions like yours. did you solve this issue?