adapt.txt

make .xml with labelImg or whatever

set folders: images/train, images/test

run xml_to_csv.py

in generate_tfrecord.py change: if row_label == '"label"':

run generate_tfrecord.py


Things to change in order to exec a different model/config/dataset/...

folders:

  - models/research/object_detection
      - data/ .record, .csv, object-detection.pbtxt     # folder with the records and cvs generated and the labels of the objects
      - "model_folder"/			      		# add folder with the downloaded model
      - images/test,train (image + xml)	     		# add folder with the images and xml annotations
      - training/"model".config             		# model config
      - export_inference_graph.py -> "folder/output"/     # folder to save the frozen inference graph after the training is done
      - test_images/ 			      		# add images to test on the jupyter notebook

scripts:

  - aux/folder/generate_tfrecord.py -> class_text_to_int -> row_label = "xxx" return 1, else if row_label == "yyy" return 2 # set accordingly to the dictionary

  - models/research/object_detection/training/"model".config:    # change the model configuration
      - num_classes: "xx"
      - batch_size: "xx"
      - fine_tune_checkpoint: "model"/model.ckpt

      - train_input_reader
          input_path: "PATH_TO_BE_CONFIGURED/mscoco_train.record-?????-of-00100"
          label_map_path: "PATH_TO_BE_CONFIGURED/mscoco_label_map.pbtxt"

      - eval_input_reader
          input_path: "PATH_TO_BE_CONFIGURED/mscoco_val.record-?????-of-00010"
          label_map_path: "PATH_TO_BE_CONFIGURED/mscoco_label_map.pbtxt"


  - models/research/object_detection/data/object-detection.pbtxt -> deffine dictionary



export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim

python3 train.py --logtostderr --train_dir=RUNS/multi/2_mix_40/ --pipeline_config_path=training/faster_rcnn_inception_resnet_v2_atrous_coco_2_mix_40k.config

tensorboard --logdir='training'

python3 export_inference_graph.py --input_type image_tensor --pipeline_config_path training/faster_rcnn_inception_resnet_v2_atrous_coco_2_mix_40k.config --trained_checkpoint_prefix RUNS/multi/2_mix_40/model.ckpt-40000 --output_directory RUNS/multi/2_mix_40/frozen40

python3 evaluate.py --inference_graph_dir=RUNS/multi/2_mix_40/frozen40/ --print_thr 0.1 --test_dir test_images/ --out_dir results/multi/2mix40/ --id "juntos"


   ------------------------------------
   useful tips
   ------------------------------------


  - if (ImportError: no module named 'xxxxx'):
       from: models/research -> export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim

  - if (error: failed to alocate memory):
	watch -n 1 free -n
	sudo sync && sudo sysctl -w vm.drop_caches=3

    elseif (previous did not work)
	close terminal, open new one

    else
	restart pc