Котики vs нейросеть 2. Или запускаем SqueezeNet v.1.1 на Raspberry Zero в realtime (почти)

import cv2
import numpy as np
import time
from time import sleep
import datetime as dt

cap = cv2.VideoCapture(0)

prototxt = 'models/solver.prototxt'
model = 'models/squeezenet_v1.1.caffemodel'
labels = 'models/synset_words.txt'

rows = open(labels).read().strip().split("\n")
classes = [r[r.find(" ") + 1:].split(",")[0] for r in rows]

# 
net = cv2.dnn.readNetFromCaffe(prototxt, model)

# reset FPS meter
t0 = time.time()

while(True):

    # take each frame
    _, frame = cap.read()
    hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)

    # only for the colors:
    lower_blue = np.array([0,0,0])
    upper_blue = np.array([255,255,255])

    # threshold the HSV image to get only blue colors
    mask = cv2.inRange(hsv, lower_blue, upper_blue)

    # putting frame into blob
    res = cv2.bitwise_and(frame,frame, mask= mask)
    blob = cv2.dnn.blobFromImage(res, 1, (224, 224), (104, 117, 124))

    # measuring the FPS 
    FPS = 1/(time.time() - t0)
    t0 = time.time()

    # trying to find the object
    net.setInput(blob)
    preds = net.forward()
    preds = preds.reshape((1, len(classes)))
    idxs = int(np.argsort(preds[0])[::-1][:1])

    # creating text for label
    text = "Label: {}, p = {:.2f}%, fps = {:.2f}".format(classes[idxs], preds[0][idxs] * 100, FPS)
    # putting text on frame
    cv2.putText(res, text, (5, 25), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)

    # prewiev
    cv2.imshow('res',res)

    # saving results 1-image for ~2.5 secconds
    # fname = 'pic_' + dt.datetime.now().strftime('%Y-%m-%d_%H-%M-%S') + '.jpg'
    # cv2.imwrite(fname, res)

    # exit from frame capturing
    k = cv2.waitKey(5) & 0xFF
    if k == 27:
        break

cv2.destroyAllWindows()