import data problem for training faster rcnn

i am tring to train my faster rccn on custom data attached ( images with there bounding boxes in .txt) but i can able to to it.kinly help me .

note:

i have multible images with ther .txt file

[Ganesh]

Hi,

I am unable to see any .txt files or images, could you please attach the needed files? Also please elaborate on your exact issue/error messages.

[ahmad]

@Ganesh

my problem is that i cant able to load all this files on matlab to train faster rcnn .as we know to train faster rcnn or any model we have to make structure like this

ans =

groundTruth with properties:

DataSource: [1×1 groundTruthDataSource]

LabelDefinitions: [5×5 table]

LabelData: [55×5 table]

but i can able to make it from dataset.

note:i have multiple images with .txt (conatain bounding boxes)

[Ganesh]

Hi,

Have you gone through the documentation for the structure of data to train Faster RCNN? The way I see it, you claim to have multiple images for the first column, you seem to have data on the bounding boxes for the second column, and the dataset should contain a label. Please refer to this document https://in.mathworks.com/help/vision/ref/trainfasterrcnnobjectdetector.html#bvkk009-1-trainingData:~:text=trainingData%20%E2%80%94%20Labeled%20ground%20truth

Kindly reach out if you have any challenges in structuring your dataset

[ahmad]

@Ganesh

Yes I have gone through it . But how can I load data from desktop and convert it into.mat formet that look like this groundTruth with properties: DataSource: [1×1 groundTruthDataSource] LabelDefinitions: [5×5 table] LabelData: [55×5 table]

Answer 1

[Githin George]

Hello Ahmad,

It is my understanding that you are trying to train a faster R-CNN network and would like to create a MATLAB “groundTruth” object for the same.

The “groundTruth” object is meant for use with the various “Labeler apps” available in MATLAB. You can programmatically create a “groundTruth” object or use a labeling app like the “Image Labeler App” to create labels and export them to the workspace or a file. Even though this object contains all the label information it cannot be used directly to train a faster R-CNN network as the input to the “trainFasterRCNNObjectDetector” function must be a table or datastore object. The data can be extracted from the object and used as inputs to the training function.

Please look at below documentation links related to the “groundTruth” object, and “trainFasterRCNNObjectDetector” function for more details.

https://www.mathworks.com/help/vision/ref/groundtruth.html

https://www.mathworks.com/help/vision/ug/elements-of-ground-truth-objects.html

https://www.mathworks.com/help/vision/ref/trainfasterrcnnobjectdetector.html#bvkk009-1-trainingData

I hope this helps.