# this script contains functions for automating
# the building of the right/wrong dataset
from imports import *
from detector_convnet import build_model
path_img = "/Users/JP/Documents/whale/imgs/"
path_img = "/Users/yc/Downloads/whale/imgs/"
# load model
print("... Loading Convolutional Network ...")
with open(path_img + "select/models/conv_3_3_3_20_2_0.03_0.005_250_200.pkl") as f:
model = load(f)
predict = build_model(model)
# load training fnames
train = pd.read_csv(path_img + "data/train.csv")
fnames = path_img + "raw/" + train["Image"]
# sliding window parameters
w1, h1 = 600, 400
w2, h2 = 100, 100
window_len, x_step, y_step = 100, 10, 10
nx, ny = (w1 - window_len) / x_step, (h1 - window_len) / y_step
# takes a pic file name, a predict function, and plots probs
def plot_pred(fname, predict):
t_open = time()
# slide over each w2*h2 region in steps of x_step, y_step
path_img = '/Users/JP/Documents/whale/imgs/' # path_img = '/Users/yc/Downloads/whale/imgs/' # path_img = '/home/jp/whale/imgs/' if sys.argv[1] is not None: [path1, path2] = sys.argv[1:3] else: path1 = 'select/models/conv_3_3_3_20_2_0.03_0.005_250_200.pkl' path2 = 'thumbs/models/conv_3_5_5_30_447_0.03_0.005_13_100_0.5.pkl' # build models from detector_convnet import build_model with open(path_img + path1) as f: model1 = load(f) detector = build_model(model1, viz=False) from classify_convnet import build_model with open(path_img + path2) as f: model2 = load(f) classifier = build_model(model2, viz=False) # read in data submit = pd.read_csv(path_img + 'data/submit.csv') # sliding window parameters w1,h1 = 600,400 w2,h2,d2 = 100,100,3 window_len, x_step, y_step = 100, 10, 10 nx,ny = (w1-window_len)/x_step, (h1-window_len)/y_step