Describe the bug
When I export annotation files that have prefix class name (e.g. S001-XXXX.jpg, S001 is class name), VoTT doesn't split test and train txt randomly. My export format was Pascal VOC.
My files are like below:
All images contain one object that specified by prefix of filename.
S001-01.jpg
S001-02.jpg
S002-01.jpg
S002-02.jpg
I want to train S002 class but if i chose test:train = 50:50 in export setting, I can not train S002 class.
The last files are not trained by train.txt, it will be all -1 since train image is not included.
Describe the bug When I export annotation files that have prefix class name (e.g.
S001-XXXX.jpg,S001is class name), VoTT doesn't split test and train txt randomly. My export format was Pascal VOC.My files are like below: All images contain one object that specified by prefix of filename.
I want to train S002 class but if i chose test:train = 50:50 in export setting, I can not train S002 class.
The last files are not trained by train.txt, it will be all
-1since train image is not included.To Reproduce
This is my repo that i got it https://github.com/mgoldchild/fasteners/tree/master/outputs/5class_fasteners_dataset Open it by VoTT and click export button.
Expected behavior Randomly split train and test images, which does not depend on filename order.
Screenshots
Desktop (please complete the following information):
OS: OS: Manjaro Linux x86_64 Processor: Intel i7-3667U (4) @ 3.200GHz RAM: 8 GB System Type: 64 bit
Additional context nothing