Tensorflow-object-detection instruction

1. Before running with a GPU, in addition to installing Docker, you will need: Up-to-date NVIDIA drivers for your system nvidia-docker: You can follow the installation instructions : https://github.com/NVIDIA/nvidia-docker#quick-start

2. Docker Build

3. RUN docker image. This docker can be run for training and evalutaion of model, by quickly setting environment variable - PROFILE to EVAL (for evaluation) and default is training.

   To run for training:

    docker run --runtime=nvidia \
    -e NVIDIA_VISIBLE_DEVICES=0 \ 
    -e PIPELINE_CONFIG=*.config \  
    -e NUM_CLONES=1 \  
    -it -p 8080:8080 \
    --mount type=bind,source={source_path},target=/root/models/research/training \
    tensorflow-object-detection 
   

   To run for evalutation:

    docker run --runtime =nvidia \
    -e NVIDIA_VISIBLE_DEVICES =0 \ 
    -e PIPELINE_CONFIG =*.config \  
    -e NUM_CLONES =1 \
     -e PROFILE = EVAL \
     -it -p 8080:8080 \
    --mount type=bind,source={source_path},target=/root/models/research/training \
    tensorflow-object-detection 
   

   Directory Structure for source_path source_path

      |—- pipeline.config
   
      ├── checkpoint
   
      ├── eval
   
      ├── pretrained_checkpoint
   
      └── tfrecord
   
            ├── test
   
            └── train
   

example: pipeline.config relative paths

1. fine_tune_checkpoint:"./training/pretrained_checkpoint/model.ckpt"
2. train_input_reader: {
		  tf_record_input_reader {
		    input_path: "./training/tfrecord/train/*.record
		  }

		  label_map_path: "./training/label_map.pbtxt"
		}
3. eval_input_reader: {
		  tf_record_input_reader {
		    input_path: "./training/tfrecord/test/*.record"
		  }

		  label_map_path: "./training/label_map.pbtxt"

		}