search

GuangmingZhu / Conv3D_CLSTM

Multimodal Gesture Recognition Using 3D Convolution and Convolutional LSTM
91 stars 35 forks source link

inputs.py #2

Open zhaiyuan opened 7 years ago

zhaiyuan commented 7 years ago

I see in inputs.py, there are different ways to processs data for different datasets or different modality. for example average_values is different, I want to how to get the average_values for a new different dataset. Because i compute the average RGB value for isogr_rgb training data and it is very different with [112,112,112] in inputs.py.

GuangmingZhu commented 7 years ago

112 is the one average value of three channels (RGB) of videos.

zhaiyuan commented 7 years ago

Thanks for your answer. I also compute the average value of three channels of all training videos, but it is not 112. I wanna know why the average_loss is nan after few iterations, it's the reason that i don't process the dataset correct or the bug in tensorflow with tf.nn.softmax(i found in stackoverflow). I found that if i modify the num_class 249 to 60 for NTU_RGB+D dataset, the average_loss is nan, when i don't modify the num_class the average_loss won't be nan, but the number of classes is 60 in NTU dataset.

Amysiat commented 6 years ago

@GuangmingZhu @zhaiyuan I want to use the code, I run the testingisogr.py,but i have a trouble "ValueError: Cannot feed value of shape(12,) for Tensor u 'x:0', which has shape '(12,32,112,112,3)' " besides,I run the training*.py also meet a trouble,I cannot find the log files. how to solve these problems ? Thank you !

zhaiyuan commented 6 years ago

you can mkdir the log file yourself. As I known, if you has the shape (12,32,112,112,3), but has the ValueError of shape (12,), you can check each data is (32,112,112,3) or not. Is a common error in python, i often see this error.

Amysiat commented 6 years ago

@zhaiyuan Then how do you solve this problem, I also encountered the same problem

zhaiyuan commented 6 years ago

Is very easy to solve this, if you have 12 videos, you should check the shape of each video data. The problem may be some videos not have 32 frames or not (112,112) or not RGB video. Is very easy to check the shape of an array(each video) in python.

zhaiyuan commented 6 years ago

you should reshape each data to a same shape.

Amysiat commented 6 years ago

@zhaiyuan Can I add your contact information to communication with you ?

baifei2018 commented 4 years ago

Is very easy to solve this, if you have 12 videos, you should check the shape of each video data. The problem may be some videos not have 32 frames or not (112,112) or not RGB video. Is very easy to check the shape of an array(each video) in python. Dear @Amysiat ，Have you solved this problem? I have the same problem. Could you please tell me how to solve it?thank you so much !!!