SmilingWolf
commented
1 month ago Thank you very much!
I've taken the liberty of applying some styling and formatting contributors have no way to know about (since there are no commit hooks, style guides and whatnot in this repo), but the functionality itself looks OK to me.
I'll be testing this later today and merge ASAP.
Added
Translators/TFRecord.py. This is a standalone script taking in a dataset folder of images and labels from text files, and outputting a specified split ofrecord_shards_trainandrecord_shards_val. Additionally, it will save the required dataset information such as class number, sample size, and unique tag ids in their respective jsons.Notify me if the structure of the output is incorrect for usage directly as training data for the generators or other project files.