simple_darkflow_tester

darkflow_tester.py

# [SETUP, get these:]
# built_graph/yolo.meta
# built_graph/yolo.pb
# (using # flow --model cfg/yolo.cfg --load bin/yolo.weights --savepb)

from darkflow.net.build import TFNet

import numpy
import os, fnmatch, random
import cv2

def load_model():
    options = {#"model": "cfg/yolo.cfg",
               #"load": "bin/yolo.weights",
               "pbLoad": "built_graph/yolo.pb",
               "metaLoad": "built_graph/yolo.meta",
               "threshold": 0.3,
               "gpu": 0.8}

    tfnet = TFNet(options)

    return tfnet

def convert_numpy_floats(result):
    # model result contains list of dictionaries, which have problematic data structure of numpy.float32
    # (in confidence). Lets convert these to me JSON-able
    for item in result:
        for key in item.keys():
            if isinstance(item[key], numpy.float32):
                item[key] = float(item[key])
    return result

def run_on_image(image_object, model):

    result = model.return_predict(image_object)
    result = convert_numpy_floats(result)

    return result

image_url = "https://www.arlingtondogandcat.com/imagebank/eVetSites/DogCats/052016_EVS_CatDog5.jpg"
image_path = "testimg.jpg"

import urllib.request
urllib.request.urlretrieve(image_url, image_path)

print("Downloaded the image ...")

image_object = cv2.imread(image_path)
darkflow_model = load_model()
print("Loaded Darkflow model ...")

result = run_on_image(image_object, darkflow_model)
print("Finished with the image ...")
print(result)

darkflow_tester_simpler.py

# [SETUP, get these:]
# built_graph/yolo.meta
# built_graph/yolo.pb
# (using # flow --model cfg/yolo.cfg --load bin/yolo.weights --savepb)
# testimg.jpg

from darkflow.net.build import TFNet
import cv2

options = {"pbLoad": "built_graph/yolo.pb",
               "metaLoad": "built_graph/yolo.meta",
               "threshold": 0.3,
               "gpu": 0.8}
               
model = TFNet(options)

image_path = "testimg.jpg"
image_object = cv2.imread(image_path)

result = model.return_predict(image_object)
print(result)