arpanghosh130883/Custom-Image-Segmentation-using-Tensorflow-1.x

Custom Image Segmentation using Tensorflow 1.x

1. Creation of Test Images

   • Download Images
   • Annotate imamges in .jason format using Labelme tool
   • Use the script to convert .jason anotated files to .tfrecords file
   • Store Images in seperate folder and anotated files in seperate folder

2. Download the model from Tensordflow model zoo

   • https://github.com/tensorflow/models/tree/v1.13.0
   • use the models whose outputs are Mask

3. Model Config file changes

   • Make necessary changes basis the requirement of experiment (sepcially in line 10 and line 127)
   • Line 127 referes for the checkpoint modification

4. Traing and Inferencing

   • Use belo commands for the same

   • Train

     • python train.py --logtostderr --train_dir=mask_training/ --pipeline_config_path=mask_training/mask_rcnn_inception_v2_coco.config

   • Inference

     • python export_inference_graph.py --input_type image_tensor --pipeline_config_path mask_training/mask_rcnn_inception_v2_coco.config --trained_checkpoint_prefix mask_training/model.ckpt-10 --output_directory inference_graph