Skip to content

Latest commit

 

History

History
114 lines (99 loc) · 2.89 KB

README.md

File metadata and controls

114 lines (99 loc) · 2.89 KB
Raw

Object detection API

Setup tensoflow MODELS

Install neccessary python modules

sudo apt-get install protobuf-compiler python-pil python-lxml python-tk 
conda install tensorflow-gpu==1.14
pip install Cython contextlib2 jupyter matplotlib cython pycocotools

Install tensorflow object detection API

wget https://github.com/tensorflow/models/archive/v1.13.0.zip
unzip v1.13.0.zip 
mv models-1.13.0 models
cd models/research
protoc object_detection/protos/*.proto --python_out=.
cd ../..
export PYTHONPATH=$PYTHONPATH:$PWD/models/research/:$PWD/models/research/slim

Configure Training Data

Download pretrained model to finetune

wget http://download.tensorflow.org/models/object_detection/faster_rcnn_inception_v2_coco_2018_01_28.tar.gz
tar -xvzf faster_rcnn_inception_v2_coco_2018_01_28.tar.gz 
rm faster_rcnn_inception_v2_coco_2018_01_28.tar.gz

Convert xml annotations to csv:

python xml_to_csv.py

Edit generate_tfrecord.py

def class_text_to_int(row_label):
    if row_label == 'basketball':
        return 1
    elif row_label == 'shirt':
        return 2
    else:
        None

Edit training/labelmap.pbtxt

item {
  id: 1
  name: 'basketball'
}

item {
  id: 2
  name: 'shirt'
}

Generate TFRecords:

python generate_tfrecord.py --csv_input=images/train_labels.csv --image_dir=images/train --output_path=train.record
python generate_tfrecord.py --csv_input=images/test_labels.csv --image_dir=images/test --output_path=test.record

Edit training/faster_rcnn_inception_v2_pets.config

  • Line 9. Change num_classes : 3 .

    num_classes: 3

  • Line 130. Change num_examples to the number of images in the /images/test directory.

     num_examples: 100

  • Line 116. Change num_steps of training steps.

     num_steps: 200000

Change below paths if any error comes

  • Line 106. Change fine_tune_checkpoint to:

     fine_tune_checkpoint : "./faster_rcnn_inception_v2_coco_2018_01_28/model.ckpt"

  • Lines 123 and 125. change input_path and label_map_path to:

     input_path : "./train.record"
 label_map_path: "./training/labelmap.pbtxt"

  • Lines 135 and 137. Change input_path and label_map_path to:

      input_path : "./test.record"
  label_map_path: "./training/labelmap.pbtxt"

Run the Training

Command:

python train.py --logtostderr --train_dir=training/ --pipeline_config_path=training/faster_rcnn_inception_v2_pets.config

Tensorboard command:

tensorboard --logdir=training

Export Inference Graph

Remove inference_graph folder if exists
Replace XXXX with ckpt number

command

python export_inference_graph.py --input_type image_tensor --pipeline_config_path training/faster_rcnn_inception_v2_pets.config --trained_checkpoint_prefix training/model.ckpt-XXXX --output_directory inference_graph