I'm trying to classify videos using LSTM but getting a different output, can anyone tell me what I'm doing wrong?

1 次查看（过去 30 天）

Suraj Tripathy 2018-2-11

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/382025-i-m-trying-to-classify-videos-using-lstm-but-getting-a-different-output-can-anyone-tell-me-what-i-m

To train-:

if true
  rootFolder = 'C:\New folder\Project\Action_Data\1\';
categories = {'jump','walk','run'};
imds = imageDatastore(fullfile(rootFolder, categories), 'LabelSource', 'foldernames');
tbl = countEachLabel(imds);
k=1;
for loop = 1:3
  i = tbl{loop,2};
  for lo = 1:i
  img = readimage(imds,lo);
  img = rgb2gray(img);
  m{k} = img;
  cate = tbl{loop,1};
  a(k) = cate;
  k = k+1;
  end
end
m = m';
a = a';
inputSize = 144;
outputSize = 100;
outputMode = 'last';
numClasses = 3;
layers = [ ...
  sequenceInputLayer(inputSize)
  lstmLayer(outputSize,'OutputMode',outputMode)
  fullyConnectedLayer(numClasses)
  softmaxLayer
  classificationLayer];
if true
  % code
end
maxEpochs = 50;
miniBatchSize = 50;
options = trainingOptions('sgdm', ...
  'MaxEpochs',maxEpochs, ...
  'MiniBatchSize',miniBatchSize);
if true
  % code
end
net = trainNetwork(m,a,layers,options);
end

And to test -:

if true
  rootFolder = 'C:\New folder\Project\Action_Data\1\';
categories = {'jump'};
imds = imageDatastore(fullfile(rootFolder, categories), 'LabelSource', 'foldernames');
tbl = countEachLabel(imds);
  for lo = 1:67
  img = readimage(imds,lo);
  img = rgb2gray(img);
  n{lo} = img;
  end
 % n = n';
  miniBatchSize = 19;
YPred = classify(net,n, ...
  'MiniBatchSize',miniBatchSize);
%acc = sum(YPred == YTest)./numel(YTest);
end

But I'm not getting the right result(Expected result-Jump, Test result-Walk. can anyone tell me what I'm doing wrong? P.S I'm taking the individual frames as input.