Collection View

A script to create an organized tableau from an image collection using unsupervised learning

Usage

Just pass a folder with the collection of images and it will generate one image with all thumbnails spatially organized.

Example with ~12000 images (resized)

Example with ~200 images (resized)

Install

1. Install Conda https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html
2. Clone this repository.
3. Change to repo directory: cd collectionview
4. Prepare the environment using: conda env create -f collection_view.yml
5. Activate the environment with: conda activate collection_view
6. Run the python script and pass a directory (subfolders are ok): python collection_view.py my_dir
7. The image will be created one directory above the image directory.

Options

-h, --help show this help message and exit

--width WIDTH width of final image

--height HEIGHT height of final image

-t THUMBWIDTH, --thumbwidth THUMBWIDTH width of each thumbnail in the final image

-m {vgg16,vgg19,resnet50}, --model {vgg16,vgg19,resnet50} choice of feature categorization model

-i SAVEIMAGE, --saveimage SAVEIMAGE name of image to be saved

-f SAVEFEATURES, --savefeatures SAVEFEATURES name of file to store the extracted features

-rp RANDOMSTATE_PCA, --randomstate_pca RANDOMSTATE_PCA PCA randomstate integer. repeat to obtain the same results

-rt RANDOMSTATE_TSNE, --randomstate_tsne RANDOMSTATE_TSNE TSNE randomstate integer. repeat to obtain the same results

To Do

• Load features saved in the pkl file
• Clustering
• Clustering with voronoi borders
• Clustering with size weight
• Mosaic organization instead of scatter