import data problem for training faster rcnn

Question

ahmad 2023-12-13

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2060144-import-data-problem-for-training-faster-rcnn

回答： Githin George 2023-12-18

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

4 个评论
显示 2更早的评论隐藏 2更早的评论

Ganesh 2023-12-13

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 2023-12-13

编辑：ahmad 2023-12-13

@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]

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Githin George 2023-12-18

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2060144-import-data-problem-for-training-faster-rcnn#answer_1373432

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.