/objection-detection

Object detetction on custom dataset using Tensorflow object detection API

Primary LanguagePython

objection-detection

Object detetction on custom dataset using Tensorflow object detection API

Installation

Type git clone https://github.com/tensorflow/models.git in the shell it will clone the TensorFlow API or it can be downloaded from here

Steps to be followed

  • First we create csv file for the images and xml file by using the xml_to_cs. First switch to that directory then type

python xml_to_csv.py

  • Then we have to create trf records for that run python generate_trfrecord.py

  • Then decide the model type which we have to use for training I used ssd_mobilnet_v1_coco_2017_11_17

  • Then we created training folder with object-detection.pbtx and config file

  • The after that we have to paste all these folder in model/research/object_detection folder.

  • Then switch to the legacy directory and run

python train.py --logtostderr --train_dir=training/ --pipeline_config_path=training/ssd_mobilenet_v1_pets.config

  • The training process graph can be seen on Tensorboard using command in shell

tensorboard --logdir='training'

  • After this for Inference Graph execute the command python3 export_inference_graph.py \ --input_type image_tensor \ --pipeline_config_path training/ssd_mobilenet_v1_pets.config \ --trained_checkpoint_prefix training/model.ckpt-(version of your model created) \ --output_directory fruits_inference_graph

  • Type this command when an import error occurs in model/research directory export PYTHONPATH=$PYTHONPATH:pwd:pwd/slim error like this No module named 'nets'

Testing the model

  • First paste some images in folder of test_images in models/research/object_detection/ directory label them image3,4,5,6.jpg

  • Now open the ipynb file for tutorial in the same directory and make these changes

# What model to download.
MODEL_NAME = 'fruit_inference_graph'

# Path to frozen detection graph. This is the actual model that is used for the object detection.
PATH_TO_CKPT = MODEL_NAME + '/frozen_inference_graph.pb'

# List of the strings that is used to add correct label for each box.
PATH_TO_LABELS = os.path.join('training', 'object-detection.pbtxt')

NUM_CLASSES = 1

  • Delete the code from download image section or do not run it.

  • And in Detection Section, change the TEST_IMAGE_PATHS var to:

TEST_IMAGE_PATHS = [ os.path.join(PATH_TO_TEST_IMAGES_DIR, 'image{}.jpg'.format(i)) for i in range(3, 8) ]
  • Then in Cell option menu click on "Run All"

The result will be images with bounding box

Error

Still I am getting this error and still unable to resolve it ValueError: Tensor conversion requested dtype string for Tensor with dtype float32: 'Tensor("arg0:0", shape=(), dtype=float32, device=/device:CPU:0)'