# Image Vectorizing and PixPlot

### Image Vectorization:
<hr>

All compiled with a file on my github called `image_vectorizer.py`. Also on a viewable [google drive link](https://drive.google.com/file/d/10KHZY-C2KnanV0AqZdY5ybJTZCQeDXK6/view?usp=sharing).

#### Significant parts of the file:

# For some reason, won't work without this.
import ssl
ssl._create_default_https_context = ssl._create_unverified_context

# Helper method I used to parse the image file names. You might need to adjust this.
def extract_page(elem: str):
    return int((elem.split('Res/')[1]).split('_')[0]) # Isolating ID.

# Seting up our model from ImageNet
    base = InceptionV3(include_top=True, weights='imagenet')
    model = Model(inputs=base.input, outputs=base.get_layer('avg_pool').output)
    vecs = []

# We actually create the vectors for each image.
    for path in paths: # Extracting just the title from each png.
        # read the images
        img = Image.open(path)  

        im = preprocess_input( img_to_array( img.resize((299,299)) ) )
        vec = model.predict(np.expand_dims(im, 0)).squeeze()
        vecs.append(vec)

The rest of the code is a creating the matrices based on the `cosine_similarity` of the vectors. Some basic CSV parsing. As for the dataset, all the code will work off of [this folder](https://drive.google.com/drive/folders/1frm8oDNf3qvdirPNv-p50DPKDArXmAVd?usp=share_link) of lower-resolution images. I'm not sure why I didn't use the **higher quality** photos, as that could've helped with vectorizations (but, maybe slowed down `Pixplot`? I'm not sure). Regardless, here are the [higher res images](https://drive.google.com/drive/folders/10B4Dm9UDOhXl3jiVQZl8enK1TO3fnQfs?usp=sharing).

### Pixplot

`Pixplot` can be compiled and then rerun just by starting a server. The compilation process is frustrating and annoying and uses a very old version of Python that **cannot be run on silicon Mac (M1, M2...)**. All of my efforts, including using a `x86-64 Rosetta` terminal didn't work - the hardware *fundamentally cannot* do it.

Luckily I compiled it on an old computer! The [compiled zip](https://drive.google.com/file/d/1LGK_G2gW8REsWLvfX6QvIe506YmzX0s1/view?usp=sharing) just needs to be downloaded and decompressed. (Be sure to have `Python 3.7.0`)

> Should you want to do this process again (that is, recompile everything with say, new images), there are some pretty bad instructions to follow [here](https://github.com/YaleDHLab/pix-plot).

Once `Python` is on your computer, open a terminal, `cd` into the folder (where you open and decompress that folder), and then run `python3 -m http.server 5000` to start a server with it.

The results will be viewable on [this website](http://localhost:5000/output).

You should see your data laid out like this: