视频二分类结果——3

接上一篇：https://blog.youkuaiyun.com/SPESEG/article/details/103403737

哈喽，大家好，我是小明哥，依旧是人见人爱花见花开百变大魔王小明哥GBM

结果不容乐观，不是跳舞的，甚至是有人的动作，更甚的是没有人也能识别为跳舞，这就尴尬了。一方面增加数据量，重新训练，另一方面看下其他的模型。

1-误识别结果，事实是全部视频都不是跳舞

other7
The total frames number=659,error rate =0.1032

animal1
The total frames number=676,error rate =0.0000

animal7
The total frames number=328,error rate =0.0000

other4
The total frames number=596,error rate =0.0000

animal5
The total frames number=1261,error rate =0.0000

other8
The total frames number=319,error rate =0.0000

animal4
The total frames number=1703,error rate =0.2719

other5
The total frames number=1237,error rate =0.0113

animal8
The total frames number=1150,error rate =0.0861

animal10
The total frames number=1807,error rate =0.0000

animal9
The total frames number=339,error rate =0.0206

other6
The total frames number=778,error rate =0.0077

other12
The total frames number=713,error rate =0.0645

other1
The total frames number=437,error rate =0.1854

animal3
The total frames number=1018,error rate =0.0236

other2
The total frames number=450,error rate =0.1489

other13
The total frames number=367,error rate =0.0163

other15
The total frames number=1865,error rate =0.0257

animal0
The total frames number=1617,error rate =0.0000

other14
The total frames number=454,error rate =0.7401

other0
The total frames number=656,error rate =0.0000

other16
The total frames number=1823,error rate =0.0000

other3
The total frames number=298,error rate =0.0000

animal6
The total frames number=793,error rate =0.0000

other9
The total frames number=776,error rate =0.0979

animal2
The total frames number=417,error rate =0.0000

other11
The total frames number=1703,error rate =0.0082

other10
The total frames number=682,error rate =0.0000

误识别的帧率占比最大74%，我看看是啥玩意，如下

就是个小妹妹对着镜头比了个剪刀手，卖个萌，这就是跳舞了？？？【图片源于抖音短视频，侵删】

这种应该不是，所以必须要增加有妹子的视频作为0视频。

鉴于这个服务器内存不足，要另一个服务器训练下试试。也可先试试另一个服务器的模型结果，如下：

0数据多了反而并不好，差别在于其中的脏数据[也就是UCF101的数据其实不应该有太多，我将其视为0视频]

other13
The total frames number=367,error rate =0.0300第一个模型0.0163

other12
The total frames number=713,error rate =0.0337第一个模型0.0645

other3
The total frames number=298,error rate =0.0000一样

animal4
The total frames number=1703,error rate =0.2202第一个模型0.2719

animal1
The total frames number=676,error rate =0.5651第一个模型为0.0000

other5
The total frames number=1237,error rate =0.0493第一个模型0.0113

other14
The total frames number=454,error rate =0.1806第一个模型0.7401

animal3
The total frames number=1018,error rate =0.6100第一个模型0.0236

other4
The total frames number=596,error rate =0.0084第一个模型为0.0000

animal2
The total frames number=417,error rate =0.0000一样

animal0
The total frames number=1617,error rate =0.0000一样

animal7
The total frames number=328,error rate =0.0000一样

other16
The total frames number=1823,error rate =0.0000一样

other6
The total frames number=778,error rate =0.0013第一个模型0.0077

other9
The total frames number=776,error rate =0.0773第一个模型0.0979

animal10
The total frames number=1807,error rate =0.0000一样

other1
The total frames number=437,error rate =0.2014第一个模型0.1854

other8
The total frames number=319,error rate =0.0000一样

animal5
The total frames number=1261,error rate =0.0040第一个模型为0.0000

animal8
The total frames number=1150,error rate =0.6452第一个模型0.0861

animal6
The total frames number=793,error rate =0.0000一样

other2
The total frames number=450,error rate =0.9667第一个模型0.1489

other10
The total frames number=682,error rate =0.0000一样

other11
The total frames number=1703,error rate =0.0458第一个模型0.0082

animal9
The total frames number=339,error rate =0.0383第一个模型0.0206

other7
The total frames number=659,error rate =0.1593第一个模型0.1032

other0
The total frames number=656,error rate =0.0000一样

other15
The total frames number=1865,error rate =0.0311第一个模型0.0257

所以应该去掉或者都试试吧，加些抖音的0视频。且看下次分解

 

少些偏见，多谢包容与认可。

另外有相关问题可以加入QQ群讨论，不设微信群

QQ群：868373192 

语音图像深度-学习群