person_vehicle_bike_detector

Person Vehicle Bike Detector

Information

The crossroad detection network model provides detection of 3 class objects: vehicle, pedestrian, non-vehicle (ex: bikes). This detector was trained on the data from crossroad cameras.

Setup

Prerequisites

• Ubuntu 16.04
• Python 3.6
• TensorFlow 1.13.1
• OpenVINO 2019 R1 with Python API

Installation

1. Download submodules

cd openvino_training_extensions
git submodule update --init --recommend-shallow external/cocoapi external/models

2. Create virtual environment

virtualenv venv -p python3 --prompt="(pvb)"

3. Modify venv/bin/activate to set environment variables

cat <<EOT >> venv/bin/activate
export PYTHONPATH=\$PYTHONPATH:$(git rev-parse --show-toplevel)/external/models/research
export PYTHONPATH=\$PYTHONPATH:$(git rev-parse --show-toplevel)/external/models/research/slim
. /opt/intel/openvino/bin/setupvars.sh
EOT

4. Activate virtual environment and setup OpenVINO variables

. venv/bin/activate

5. Install modules

pip3 install -r requirements.txt

6. Build and install COCO API for python

cd $(git rev-parse --show-toplevel)/external/cocoapi
2to3 . -w
cd PythonAPI
make install

7. Compile Protobuf libraries

cd $(git rev-parse --show-toplevel)/external/models/research/
protoc object_detection/protos/*.proto --python_out=.

Training and evaluation example

NOTE To train model on own dataset you should change num_steps: 10 in configs/pipeline.config.

1. Go to openvino_training_extensions/tensorflow_toolkit/veh_ped_nonveh_ssd_mobilenetv2_detector/ directory

2. The example dataset has annotation in coco format. You can find it here: openvino_training_extensions/tensorflow_toolkit/veh_ped_nonveh_ssd_mobilenetv2_detector/dataset To collect annotation used COCO object detection format. .

3. To convert the dataset to tfrecords you have to run:

   python ./tools/create_crossroad_extra_tf_records.py \
       --train_image_dir=../../data/airport/train/ \
       --val_image_dir=../../data/airport/val/ \
       --train_annotations_file=../../data/airport/annotation_example_train.json \
       --val_annotations_file=../../data/airport/annotation_example_val.json \
       --output_dir=../../data/airport/tfrecords

4. To start training you have to run:

   python ../../external/models/research/object_detection/model_main.py \
     --model_dir=./model \
     --pipeline_config_path=./configs/pipeline.config

   Training artifacts will be stored by default in model

5. Evalution artifacts will be stored by default in openvino_training_extensions/tensorflow_toolkit/veh_ped_nonveh_ssd_mobilenetv2_detector/model/eval_0/. To show results of network model working run

   tensorboard --logdir=./model

   And view results in a browser: http://localhost:6006.

Freezing graph

python ../../external/models/research/object_detection/export_inference_graph.py \
  --input_type=image_tensor \
  --pipeline_config_path=./configs/pipeline.config \
  --trained_checkpoint_prefix=./model/model.ckpt-10 \
  --output_directory ./model/export_10

Demo

python tools/infer.py --model=model/export_10/frozen_inference_graph.pb \
  --label_map=../../data/airport/crossroad_label_map.pbtxt \
  ../../data/airport/val/image_000000.jpg

Conversion to Intermediate Representation (IR) of the network

"${INTEL_OPENVINO_DIR}"/deployment_tools/model_optimizer/mo_tf.py \
  --model_name veh_ped_nonveh_ssd_mobilenetv2_detector \
  --input_model=./model/export_10/frozen_inference_graph.pb \
  --output_dir=./model/export_10/IR \
  --tensorflow_object_detection_api_pipeline_config=./configs/pipeline.config \
  --tensorflow_use_custom_operations_config="${INTEL_OPENVINO_DIR}/deployment_tools/model_optimizer/extensions/front/tf/ssd_v2_support.json"

Demo

python tools/infer_ie.py --model model/export_10/IR/veh_ped_nonveh_ssd_mobilenetv2_detector.xml \
  --device=CPU \
  --cpu_extension="${INTEL_OPENVINO_DIR}/deployment_tools/inference_engine/lib/intel64/libcpu_extension_avx2.so" \
  --label_map dataset/crossroad_label_map.pbtxt \
  dataset/ssd_mbv2_data_val/image_000000.jpg