Count Object Scale

Count-Object-Scale.md

from data import IceSegmentation
from model import PhaseFourierTransform
import numpy as np
import scipy as sp
import matplotlib.pyplot as plt
import platform, os
from skimage import measure, color
from mxnet import nd

#######################################################
################## Hyper-Parameter ####################
#######################################################
sal_size = 64
orig_size = 512
std_coef = 12.5
kernel_size = 5

show_flag = False

if platform.system() == "Darwin":
    data_root = os.path.join('~', 'Nutstore Files', 'Dataset', 'Iceberg')
elif platform.system() == "Linux":
    data_root = os.path.join('~', 'datasets', 'Iceberg')
else:
    raise ValueError('Notice Dataset Path')


train_dataset = IceSegmentation(split="trainvaltest", mode='testval', base_size=512, crop_size=512)
mean_mat = nd.zeros((len(train_dataset), 3))
std_mat = nd.zeros((len(train_dataset), 3))

hist_arr = np.zeros(512)
for i, data in enumerate(train_dataset):
    # img = data[0]
    # mean_mat[i, :] = data[0].mean(axis=(0,1))
    # std_mat[i, :] = data[0].std(axis=(0,1))
    # print("begin")
    # print(data[1].max())
    labels = measure.label(data[1].asnumpy(), background=0)
    # print("end")
    # print(labels)
    # print(data[2])
    for i, region in enumerate(measure.regionprops(labels)):
        minr, minc, maxr, maxc = region.bbox
        height = maxr - minr + 1
        width = maxc - minc + 1
        scale = min(height, width)
        hist_arr[scale-1] += 1
        # print("BBox ", i, ", Scale: ", scale)
        # print("minr: ", minr)
        # print("minc: ", minc)
        # print("maxr: ", maxr)
        # print("maxc: ", maxc)

    # break
print(hist_arr)

# from tempfile import TemporaryFile
# hist_arr_file = TemporaryFile()
np.save("hist_arr", hist_arr)
# hist_arr_file.seek(0)
# np.load(hist_arr_file)
# print("Mean Values: ", mean_mat.mean(axis=0))
# print("Std Values: ", std_mat.mean(axis=0))