purrsong

Installation

pip install purrsong

Start

import purrsong as ps
ps.__version__
>>> 0.1.6

Create Anaconda Environment(Optional)

conda create -n purrsong python=3.6
conda activate purrsong

Requirments

• tensorflow
• opencv-python
• tqdm
• requests
• pandas
• matplotlib

Auto-download-extract-load of datasets, modelsets

Below function automatically download data or models and save locally If data is already exists, returns data directory or model filepath

import purrsong as ps

model_list = ps.list_models() # or ps.list_models(fresh=True)
print(model_list)
ps.load_model('bbs') # or ps.load_model('bbs', fresh=True)

dataset_list = ps.list_datasets() # or ps.list_datasets(fresh=True)
print(dataset_list)
ps.load_dataset('cat') # or ps.load_dataset('cat', fresh=True)

Manipulating cats dataset

You can play with auto downloaded cats dataset by below example. Try changing factor arg, which will return different size of bounding boxes.

load cats dataset

import purrsong as ps
import matplotlib.pyplot as plt
cats = ps.load_cats()

showing cat image

img = cats[0]['image']
plt.imshow(img)
plt.show()

showing cat image with landmark

img = cats[0]['image']
lmk = cats[0]['landmark']
x, y = lmk.T
plt.imshow(img)
plt.scatter(x, y)
plt.show()

showing cat face image

img = cats.face_img(44)  # or img = cats.face_img(idx=44, factor=1.7)
plt.imshow(img)
plt.show()

showing cat left eye image

img = cats.left_eye_img(44)  # or img = cats.face_img(idx=44, factor=1.7)
plt.imshow(img)
plt.show()

available data features

import purrsong as ps
cats = ps.load_cats()

cat = cats(0)         # or cats(idx=0, factor=1.7)

cat['image']          # cats.image(0)
cat['landmark']       # cats.landmark(0)
cat['face']           # cats.face(0, factor=1.7)
cat['face_bb']        # cats.face_bb(0, factor=1.7)
cat['face_img']       # cats.face_img(0, factor=1.7)
cat['face_lmk']       # cats.face_lmk(0, factor=1.7)
cat['eye']            # cats.eye(0, factor=1.7)
cat['left_eye_bb']    # cats.left_eye_bb(0, factor=1.7)
cat['left_eye_img']   # cats.left_eye_img(0, factor=1.7)
cat['right_eye_bb']   # cats.right_eye_bb(0, factor=1.7)
cat['right_eye_img']  # cats.right_eye_img(0, factor=1.7)
cat['nose']           # cats.nose(0, factor=1.7)
cat['nose_bb']        # cats.nose_bb(0, factor=1.7)
cat['nose_img']       # cats.nose_img(0, factor=1.7)
cat['ear']            # cats.ear(0, factor=1.7)
cat['left_ear_bb']    # cats.left_ear_bb(0, factor=1.7)
cat['left_ear_img']   # cats.left_ear_img(0, factor=1.7)
cat['right_ear_bb']   # cats.right_ear_bb(0, factor=1.7)
cat['right_ear_img']  # cats.right_ear_img(0, factor=1.7)

left dict form is much more intuitive and good when you have to handle many of features at the same time.
right method way is good when you access specific feature.