aakaar

Installation instructions

• Install python3 and virtualenv if it is not installed for your platform

For Ubuntu: sudo apt-get install virtualenv

For OSX: brew install virtualenv

• Clone this repository

git clone [email protected]:sahajsoft/aakaar.git

• Inside aakaar directory create a virtualenv with python3

cd aakaar/ virtualenv -p python3 venv

• Source virtualenv

`source venv/bin/activate~

• Install all required pip dependencies

pip install -r requirements.txt

• Outside aakaar directory clone tensorflow models repo

git clone https://github.com/tensorflow/models

• In your shell rc file import the path for the slim form models/research directory

vi ~/.bashrc or vi ~/.zshrc depending on the shell you are using Add line at the end export PYTHONPATH=$PYTHONPATH:<Directory_To_Models_Parent_Folder>/models/research:<Directory_To_Models_Folder>/models/research/slim

• Git clone cocoapi and make

git clone https://github.com/cocodataset/cocoapi.git cd cocoapi/PythonAPI make

• Copy pycoco tools to the models

cp -r pycocotools <Directory_To_Models_Parent_Folder>/models/research/

• Generate protobuf files from the config

Ensure that correct version of protobuf is installed. I had 3.5+ installed.

cd <Directory_To_Models_Parent_Folder>/models/research/ protoc object_detection/protos/*.proto --python_out=.

• Run simple tensorflow test to ensure everything is correctly installed

cd <Directory_To_Models_Parent_Folder>/models/research python object_detection/builders/model_builder_test.py

Execution

Local Machine

PIPELINE_CONFIG_PATH="/Users/priyank/Projects/aakaar/model/config"
MODEL_DIR="/Users/priyank/Projects/aakaar/model/"
NUM_TRAIN_STEPS=50000
SAMPLE_1_OF_N_EVAL_EXAMPLES=1
python object_detection/model_main.py \
    --pipeline_config_path=${PIPELINE_CONFIG_PATH} \
    --model_dir=${MODEL_DIR} \
    --num_train_steps=${NUM_TRAIN_STEPS} \
    --sample_1_of_n_eval_examples=$SAMPLE_1_OF_N_EVAL_EXAMPLES \
    --alsologtostderr

AWS GPU Instance

PIPELINE_CONFIG_PATH="/home/ubuntu/projects/aakaar/model/config.server"
MODEL_DIR="/home/ubuntu/projects/aakaar/model/"
NUM_TRAIN_STEPS=50000 
SAMPLE_1_OF_N_EVAL_EXAMPLES=1
python object_detection/model_main.py \
    --pipeline_config_path=${PIPELINE_CONFIG_PATH} \    
    --model_dir=${MODEL_DIR} \
    --num_train_steps=${NUM_TRAIN_STEPS} \
    --sample_1_of_n_eval_examples=$SAMPLE_1_OF_N_EVAL_EXAMPLES \
    --alsologtostderr

Exporting frozen reference graph

python /Users/priyank/PetProjects/models/research/object_detection/export_inference_graph.py --input_type image_tensor --pipeline_config_path ~/Projects/aakaar/model/config --trained_checkpoint_prefix ~/Projects/aakaar/model/model.ckpt-50000.data-00000-of-00001 --output_directory model_output