carlosgonzalez / image-tools

Tools for creating image-based datasets for machine learning

Image Tools: creating image datasets

Image Tools helps you form machine learning datasets for image classification.

Download the desktop application on Windows

We use GitHub Actions to build the desktop version of this app. If you would like to download it for Windows, please click on Actions and then you will see PyInstaller Windows on the left under 'All Workflows'. Once you click into the Windows workflow, you will see a list of the builds in the center of the screen. Click on the topmost item in the results list for the latest version. Once you click on the latest build, you should see a section titled 'Artifacts' with an item called 'Image Tools Windows'. When you click on this artifact, it should download the zip containing the app for you!

Support for MacOS is still in progress.

Run Desktop application on MacOS

While the compiled Mac app does not work because it is unsigned, you can either create it locally or run the app via Python command line.

1. Make sure you have Python 3.8 or Python 3.9. See guide here for installing: https://www.python.org/downloads/mac-osx/
2. Download this code repository to your machine.
3. Open a terminal and navigate to where you downloaded this code with the cd command, or open the terminal directly at the Image Tools folder by right clicking on it and selecting Services > Open Terminal at Folder
4. Install dependencies via pip3 install -r requirements.txt (make sure your pip3 is pointing to your Python 3.8 or 3.9 installation by checking pip3 --version)
5. Run the app!
   • Run from command line: python3 -m app.app
   • Compile the app yourself: pyinstaller --windowed --onefile app/app.spec. This will create a dist/ folder that contains the app you can run.

Code Setup

Install the required packages.

pip install -r requirements.txt

If you are on Windows, you will also need to install the latest PyInstaller from GitHub:

pip install git+https://github.com/pyinstaller/pyinstaller.git

Command Line Usage

CSV, XLSX, or TXT files

Downloading an image dataset from the urls in a csv or txt file:

python -m dataset.download_from_file your_file.csv --url UrlHeader --label LabelHeader

This downloader script takes either a csv, xlsx, or txt file and will format an image dataset for you. The resulting images will be downloaded to a folder with the same name as your input file. If you supplied labels, the images will be grouped into sub-folders with the label name.

• csv or xlsx file

  • specify the column header for the image urls with the --url flag
  • you can optionally give the column header for labels to assign the images if this is a pre-labeled dataset

• txt file

  • separate each image url by a newline

Predicting labels and confidences for images in a csv, xlsx, or txt file:

python -m model.predict_from_file your_file.csv path/to/lobe/savedmodel --url UrlHeader

This prediction script will take a csv or txt file with urls to images and a Lobe TensorFlow SavedModel export directory, and create and output csv with the label and confidence as the last two columns.

• csv or xlsx file

  • specify the column header for the image urls with the --url flag

• txt file

  • separate each image url by a newline

Folder of images

python -m model.predict_from_folder path/to/images path/to/lobe/savedmodel

This prediction script will take a directory of images and a Lobe TensorFlow SavedModel export directory, and reorganize those images into subdirectories by their predicted label.

Flickr downloader

Download images from Flickr by latitude and longitude bounding box location and any desired search terms.

python -m dataset.download_from_flickr api_key dest_folder --bbox min_lat,min_long,max_lat,max_long --search searchTerm

This will create an images.csv file in your destination folder that includes the EXIF data for the downloaded photos.

Export Lobe dataset

Export your project's dataset by giving the project name and desired export directory:

python -m dataset.export_from_lobe 'Project Name' destination/export/folder

Your images will be copied to the destination folder, and their labels will be the subfolder name. You can take this exported folder and drag it directly to a new project in Lobe.

Build Desktop Application

You can create a desktop GUI application using PyInstaller:

pyinstaller --onefile --windowed app/app.spec

This will create a dist/ folder that will contain the application file Image Tools.exe or Image Tools.app depending on your OS.

Running the desktop application for development

python -m app.app