Deep Learning: How do I add regression ground truth data to an imageDatastore?

if I train my network with newnet=trainNetwork(imageDatastore_traindata,net,options) imageDatastore_traindata.Labels need to be categorical, i.e. not regression but classification labels. So for regression, it seems I need to use newnet=trainNetwork(X,Y,net,options)

and give my regression labels as Y, but then I can't use the imageDatastore as X, so I'll run into memory problems. Is there no way to do regression with an imageDatastore?

[Bashar Saad]

hi friend i have same the problem i do not know how to load my own dataset in deep learning could you help me plase ?

Answer 1

[Amy]

You can use an image datastore to get the image file names (without having to load the images from the dataset into memory at all), then create a table with the file names in the first column and your regressors in the other columns, and feed that into trainNetwork.

(see the trainedNet = trainNetwork(tbl,layers,options) syntax for trainNetwork.)

[Christoph Block]

Is there a similar way for non image matrices? e.g. 3Dmultichannel matrix stored in a .mat-file? It seems that a table with paths in the first column requires jpg or png files.

[Dammak]

Thank you Amy!

is there a way to still apply the options available in augmentedImageDatastore when using this table format?

[Tomer Nahshon]

Joins Christoph Block on the question.

any answer?

[Simone Scaringi]

Also joining Christoph Block with same question: i.e. is there a way to parse the tbl argument and allow the trainNetwork function to read files using e.g. a custom ReadFcn ?

[Simone Scaringi]

I've spent way too much time getting my head around this.

What I ended up doing to "fix" it, is to save my data files as .png images. In my case these are just 1D "images", but I guess you can save the 3D matrix as an image as well. If you do this then train proceeds as expected since the table with file paths have .png "images" in them.

Hope this helps to whoever reads this!

Answer 2

[Daniel Cohen]

% trainedNet = trainNetwork(ds, layers, options) trains and returns a

% network trainedNet using the datastore ds. For single-input networks,

% the datastore read function must return a two-column table or

% two-column cell array, where the first column specifies the inputs to

% the network and the second column specifies the expected responses. For

% networks with multiple inputs, the datastore read function must return

% a cell array with N+1 columns, where N is the number of inputs. The

% first N columns correspond to the N inputs and the final column

% corresponds to the responses.

What what does that mean for this situation?