-
research 폴더에서
protoc object_detection/protos/*.proto --python_out=.
export PYTHONPATH=$PYTHONPATH:pwd:pwd/slim
export PYTHONPATH=$PYTHONPATH:/path/to/models
python setup.py build
python setup.py install
-
research/object_detection/builders 경로에서
python model_builder_test.py
-
object_detection 폴더에서
PIPELINE_CONFIG_PATH=/home/john/DLstorage/repos/models/research/object_detection/samples/configs/faster_rcnn_resnet101_coco.config
MODEL_DIR=/home/john/DLstorage/result
NUM_TRAIN_STEPS=10000000
python model_main_tf2.py --pipeline_config_path=${PIPELINE_CONFIG_PATH} --model_dir=${MODEL_DIR} --num_train_steps=${NUM_TRAIN_STEPS} --alsologtostderr
python legacy/train.py --pipeline_config_path=/home/john/project/repos/models/research/object_detection/samples/configs/faster_rcnn_resnet101_coco.config --train_dir=/home/john/DLstorage/result --logtostderr --worker_replicas=1 --num_clones=1 --ps_tasks=1
(example)
python generate_tfrecord.py --csv_input=/home/john/DLstorage/repos/models/research/object_detection/images/images/train_labels.csv --image_dir=/home/john/DLstorage/repos/models/research/object_detection/images/images/train --output_path=/home/john/DLstorage/train.record
python legacy/eval.py --checkpoint_dir=/home/john/DLstorage/result_jpg --eval_dir=./test_images --pipeline_config_path=./samples/configs/faster_rcnn_resnet101_coco.config
https://medium.com/analytics-vidhya/training-an-object-detection-model-with-tensorflow-api-using-google-colab-4f9a688d5e8b <- useful
https://medium.com/@shubham.borikar/train-tensorflow-object-detection-api-on-custom-dataset-839ebb93dddc <- useful
2020 ver
https://neptune.ai/blog/how-to-train-your-own-object-detector-using-tensorflow-object-detection-api
https://medium.com/swlh/tensorflow-2-object-detection-api-with-google-colab-b2af171e81cc
https://medium.com/@deep12vish/tensorflow-object-detection-in-windows-under-30-lines-d6776586c4ab