after i followed the hello AI World example for training ssd object detection i wanted to try training it on my dataset for scorpions. but the openimages.com site doesn’t have enough data for it so i created my own and used " .CSV "format since the example were using it.
the problem is that i don’t know how to add it the directory and what hierarchy should it be
Hi @BLACK_CAT, if you are using CSV, it should mimic the hierarchy of the Open Images set when it’s downloaded. However for custom dataset, it may be easier to just use Pascal VOC format as that has a simpler structure. And you can use the CVAT tool to annotate your images and export the dataset in Pascal VOC format.
Then after adding a labels.txt file, you can train it like this: https://github.com/dusty-nv/jetson-inference/blob/master/docs/pytorch-collect-detection.md#training-your-model
-the original file has tow CSV file for train and test (train-annotations-bbox and sup-train-annotations-bbox ) so i got confused and even if i tried to copy the sup flies i couldnt
-i have also tried it with Pascal VOC 1.1 and got this error.
Hi @BLACK_CAT, sorry for the delay - for Pascal VOC, you need these files under ImageSets/Main:
test.txt train.txt trainval.txt val.txt
If you don’t have them all, you can copy the ones that you do have.
hey thanks for the replay and sorry for the delay.
how can i great these files lists ?
Hi @BLACK_CAT, do you have at least one of the above file lists? If so, you can copy one of them to the others.
sorry for my English what i mean is what are the lists for ? do i have to put the images names in these files ?
The lists contain the image IDs. For example, the first 5 lines of my train.txt look like this:
20200916-191931
20200916-192013
20200916-192033
20200916-192107
20200916-192122
These correspond to the image filenames under JPEGImages/ (i.e. 20200916-191931.jpg) and the XML annotations under Annotations/ (i.e. 20200916-191931.xml)
This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.