Transform my own datasets (Similar to Pascalvoc2007)dataset format to tfrecord
XN-0808 opened this issue · 1 comments
XN-0808 commented
There is a problem when i tried to convert my own datasets to tfrecord fromat
my datasets is similar to PascalVOC2007,the difference is that the ImageSets/Main only contains four files:"train.txt","val.txt","trainval.txt","test.txt",
i don't know how to revise the "pascalvoc.py",can someone help me?
kshitijagrwl commented
You can have a look at the basic tutorial I've posted here
For the pascalvoc reader, you need to modify the following:
get_classes()
that dervices from the files specified under ImageSets/Main_get_record_names()
that generates each image_id to be read and depends on_get_split_path()
Your test.txt, train.txt etc files should follow the pascalvoc convention and it should be fine!