Flo²Plan Dataset Manager

Florence Floor Plan is an image dataset for ORF competition (). This tool is useful to manage the dataset and evaluate the results.

Requirements

• Anaconda 3
• COCO Tools

Installation

Anaconda

You need install Anaconda, then run below:

# python3

conda create --name flo2 python=3.6
source activate flo2
#pycocotools (https://github.com/cocodataset/cocoapi/tree/master/PythonAPI)
conda install -c hcc pycocotools

Usage

Download images

You can find the list of image files on the competion website as a JSON file. Having that JSON file, it is possible to download the imageset using the next command:

#command
./download_dataset.py <json_file> --dir <local_dir>

In this repository you can find one JSON file (an example) to try the command.

#run the example:
./download_dataset.py example_images.json --dir .

In this way you can download the images and create the dataset directories inside your local path. In your local path you will find a directory named Flo2Plan:

# list of files
./Flo2Plan
./Flo2Plan/Images
./Flo2Plan/Images/G0375.png
./Flo2Plan/Images/G0423.png
./Flo2Plan/Images/G0592.png
./Flo2Plan/output.json

The file named output.json contains all the information about each downloaded images

#output.json
[
  {
    "key": "G0592",
    "image_mode": "RGB",
    "result": "OK"
  },
  {
    "key": "G0375",
    "image_mode": "RGB",
    "result": "OK"
  },
  {
    "key": "G0423",
    "image_mode": "P",
    "result": "OK"
  }
]

Each file has a key, the downloading result (result) and other information related to the original URL image path. The values for result are OK or KO. If you find some items where result is KO this means you cannot download the related file from the specific URL. In this case, please contact us.

Visualize Instances

It is possible to visualize the instances on image using the following script:

#visualize
./visual_instances_image.py <instances_json_file> --img_dir <image_dir_path> --key <image_key>

The key_img is the image key which you can find inside the JSON file to download files (example: example_image.json). If you use the key_img as ALL you can create a local directory where the script will save the overlayed image with instance masks. Otherwise, if key_img is empy, the script will sample a random key from the ground truth.

#random image from dataset
./visual_instances_image.py example_instances.json --img_dir Flo2Plan/trainingset/images

#image with key G0423
/visual_instances_image.py example_instances.json --img_dir Flo2Plan/trainingset/images --key G0423

#salve all the overlayed images in local directory
/visual_instances_image.py example_instances.json --img_dir Flo2Plan/trainingset/images --key ALL --save_path local

Evaluate results
In order to evaluate your results, you can use the following script:

#evaluate
./evaluation_results.py ....... ??? ......

coming soon