Skip to content

Latest commit

 

History

History
63 lines (51 loc) · 3.18 KB

getting-started.md

File metadata and controls

63 lines (51 loc) · 3.18 KB
Raw

Getting Started with COCO Synth

Initial Setup

I highly recommend using Anaconda for Python environment management. It will help you install Shapely, which I've had some problems installing with pip.

Once you have Anaconda installed...

On Windows:

conda create -n cocosynth python=3.6
activate cocosynth
conda install -c conda-forge shapely
pip install -r requirements.txt

On Linux (and I assume Mac):

conda create -n cocosynth python=3.6
source activate cocosynth
conda install -c conda-forge shapely
pip install -r requirements.txt

View Segmentations with the COCO Image Viewer

Fire up Jupyter Notebook.

jupyter notebook

Open up "../notebooks/coco_image_viewer.ipynb" and run through the cells in the notebook. Pay attention to the file paths. They are set up to work with this guide. If everything works correctly, you'll be able to view an image with image segmentation overlays.

Create Synthetic Images and Masks

In this section, we will use "image_composition.py" to randomly pick foregrounds and automatically super-impose them on backgrounds. You will need a number of foreground cutouts with transparent backgrounds. For example, you might have a picture of an eagle with a completely transparent background. Due to the need for transparency, these images should be .png format (.jpg doesn't have transparency). I cut out my foregrounds with GIMP, which is free.

Using the sample dataset

For this guide, all examples assume you'll be using the box_dataset_synthetic sample dataset. Find it here ../datasets/README.md. Download it and extract the contents to "../datasets/box_dataset_synthetic".

Custom dataset directory setup:

  • Inside the "datasets" directory, create a new folder for your dataset (e.g. "wild_animal_dataset")
  • Inside that dataset directory, create a folder called "input"
  • Inside "input", create two folders called "foregrounds" and "backgrounds"
  • Inside "foregrounds", create a folder for each super category (e.g. "bird", "lizard")
  • Inside each foreground super category folder, create a folder for each category (e.g. "eagle", "owl")
  • Inside each category folder, add all foreground photos you intend to use for the respective category (e.g. all of you eagle foreground cutouts)
  • Inside "backgrounds", add all background photos you intend to use

Run "image_composition.py" to create your images and masks

python ./python/image_composition.py --input_dir ./datasets/box_dataset_synthetic/input --output_dir ./datasets/box_dataset_synthetic/output --count 10 --width 512 --height 512

Create COCO Instances JSON

Now we're going to use the images, masks, and json to create COCO instances.

Optional: Run "coco_json_utils.py" with --help to see the documentation. This will explain the next command.

python ./python/coco_json_utils.py --help

Run the command with the correct parameters

python ./python/coco_json_utils.py -md ./datasets/box_dataset_synthetic/output/mask_definitions.json -di ./datasets/box_dataset_synthetic/output/dataset_info.json

You will now have a new json file called "coco_instances.json". This is contains all of your COCO json!