existedinnettw / riceCV

understand google object detection API

在 Model Garden garden 裡大致有分official 和 research ，official 裡的object detection API 似乎主要是從頭開始train（還不確定model 是否會根據input image 而變），是直接用python 去執行script。由於我需要fine tune...，等多變化，自己設定執行步驟，所以不是用official 裡的東西

research 裡有各種model （Model Zoo），執行到一半的checkpoint，model config。適合做各種train 方法，步驟的更換，以jupyter notebook 為主，這才是我要用的。

• Object Detection API 2.0, error with load checkpoints: A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. #8892

  • finetune type 設detection 而非classification

• “tensorflow.python.framework.errors_impl.FailedPreconditionError” while running “model_main_tf2.py” for training object detection model in tensorflow

  • model目錄內不要有checkpoint 資料夾

​

train

python3 "/mnt/c/Users/insleker/Google Drive/workspace/riceCV/models/research/object_detection/model_main_tf2.py" --pipeline_config_path=./pipeline.config --model_dir=. --alsologtostderr

eval

python3 "/mnt/c/Users/insleker/Google Drive/workspace/riceCV/models/research/object_detection/model_main_tf2.py" --pipeline_config_path=./pipeline.config --model_dir=. --checkpoint_dir=. --alsologtostderr

然後開tensorboar

tensorboard --logdir=.

---

export

TensorFlow Object Detection API: Best Practices to Training, Evaluation & Deployment

python3 "/mnt/c/Users/insleker/Google Drive/workspace/riceCV/models/research/object_detection/exporter_main_v2.py" --pipeline_config_path=./pipeline.config --trained_checkpoint_dir=. --output_directory="../exported_models/?" --input_type=image_tensor