Custom Dataset to tfrecord
Apple M1 python 3.10 tensorflow 2.18
I used "LabelImg" tool
You should change data structure format like "PASCAL VOC2012"
📦VOC2012
┣ 📂Annotations
┃ ┣ 📜sample.xml # tag must same JPEGImages's filename
┃ ┗ 📜sample2.xml
┣ 📂ImageSets
┃ ┗ 📂Main
┃ ┃ ┣ 📜Mold_train.txt # change class name into your dataset class information
┃ ┃ ┗ 📜Mold_val.txt
┣ 📂images
┃ ┗ 📂JPEGImages
┃ ┃ ┣ 📜sample.jpg
┃ ┃ ┗ 📜sample2.jpg
┗ 📜label_map.json # label information, check name and number
!python functions/create_pascal_tfrecord.py
--data_dir="Dataset/VOC2012"
--set=train
--year=VOC2012
--annotations_dir="Annotations"
--output_path="results/results"
--num_shards=10
--label_map_json_path="Dataset/VOC2012/label_map.json"
if sucess, results folder have tfrecords and .json file
** functions from https://github.com/google/automl