import numpy as np
import pandas as pd
from scipy.misc import imread
import cv2
import os
from scipy import ndimage as ndi
from skimage.morphology import remove_small_objects, watershed, remove_small_holes
from skimage import measure
# Run-length encoding stolen from https://www.kaggle.com/rakhlin/fast-run-length-encoding-python
def rle_encoding(x):
dots = np.where(x.T.flatten() == 1)[0]
run_lengths = []
prev = -2
for b in dots:
if (b>prev+1): run_lengths.extend((b + 1, 0))
run_lengths[-1] += 1
prev = b
return run_lengths
def prob_to_rles(lab_img):
# lab_img = label(x > cutoff)
for i in range(1, lab_img.max() + 1):
yield rle_encoding(lab_img == i)
def my_watershed(what, mask1, mask2):
# markers = ndi.label(mask2, output=np.uint32)[0]
# big_seeds = watershed(what, markers, mask=mask1, watershed_line=False)
# m2 = mask1 - (big_seeds > 0)
# mask2 = mask2 | m2
markers = ndi.label(mask2, output=np.uint32)[0]
labels = watershed(what, markers, mask=mask1, watershed_line=True)
# labels = watershed(what, markers, mask=mask1, watershed_line=False)
return labels
def calc_score(labels, y_pred):
true_objects = len(np.unique(labels))
pred_objects = len(np.unique(y_pred))
# print("Number of true objects:", true_objects)
# print("Number of predicted objects:", pred_objects)
# Compute intersection between all objects
intersection = np.histogram2d(labels.flatten(), y_pred.flatten(), bins=(true_objects, pred_objects))[0]
# Compute areas (needed for finding the union between all objects)
area_true = np.histogram(labels, bins=true_objects)[0]
area_pred = np.histogram(y_pred, bins=pred_objects)[0]
area_true = np.expand_dims(area_true, -1)
area_pred = np.expand_dims(area_pred, 0)
# Compute union
union = area_true + area_pred - intersection
# Exclude background from the analysis
intersection = intersection[1:, 1:]
union = union[1:, 1:]
union[union == 0] = 1e-9
# Compute the intersection over union
iou = intersection / union
# Precision helper function
def precision_at(threshold, iou):
matches = iou > threshold
true_positives = np.sum(matches, axis=1) == 1 # Correct objects
false_positives = np.sum(matches, axis=0) == 0 # Missed objects
false_negatives = np.sum(matches, axis=1) == 0 # Extra objects
tp, fp, fn = np.sum(true_positives), np.sum(false_positives), np.sum(false_negatives)
return tp, fp, fn
# Loop over IoU thresholds
prec = []
# print("Thresh\tTP\tFP\tFN\tPrec.")
for t in np.arange(0.5, 1.0, 0.05):
tp, fp, fn = precision_at(t, iou)
p = tp / (tp + fp + fn)
# print("{:1.3f}\t{}\t{}\t{}\t{:1.3f}".format(t, tp, fp, fn, p))
prec.append(p)
# print("AP\t-\t-\t-\t{:1.3f}".format(np.mean(prec)))
return np.mean(prec), prec
def wsh(mask_img, threshold, border_img, seeds):
img_copy = np.copy(mask_img)
m = seeds * border_img# * dt
img_copy[m <= threshold + 0.35] = 0
img_copy[m > threshold + 0.35] = 1
img_copy = img_copy.astype(np.bool)
img_copy = remove_small_objects(img_copy, 10).astype(np.uint8)
mask_img[mask_img <= threshold] = 0
mask_img[mask_img > threshold] = 1
mask_img = mask_img.astype(np.bool)
mask_img = remove_small_holes(mask_img, 1000)
mask_img = remove_small_objects(mask_img, 8).astype(np.uint8)
# cv2.imwrite('t.png', (mask_img * 255).astype(np.uint8))
# cv2.imwrite('t2.png', (img_copy * 255).astype(np.uint8))
labeled_array = my_watershed(mask_img, mask_img, img_copy)
return labeled_array
def postprocess_victor(pred):
av_pred = pred / 255.
av_pred = av_pred[..., 2] * (1 - av_pred[..., 1])
av_pred = 1 * (av_pred > 0.5)
av_pred = av_pred.astype(np.uint8)
y_pred = measure.label(av_pred, neighbors=8, background=0)
props = measure.regionprops(y_pred)
for i in range(len(props)):
if props[i].area < 12:
y_pred[y_pred == i + 1] = 0
y_pred = measure.label(y_pred, neighbors=8, background=0)
nucl_msk = (255 - pred[..., 2])
nucl_msk = nucl_msk.astype('uint8')
y_pred = watershed(nucl_msk, y_pred, mask=((pred[..., 2] > 80)), watershed_line=True)
return y_pred
# test_dir = r'C:\dev\dsbowl\results_test\bowl_remap3\merged'
# borders_dir = r'C:\dev\dsbowl\results_test\bowl_remap_border2\merged'
oof = True
# borders_dir = r'd:\tmp\bowl\results_test\s4b\merged'
if oof:
# test_dir = r'/home/albu/dev/bowl/results/dpn_sigm_3channel'
test_dir = r'c:\dev\dsbowl\results\dpn_sigm_f0'
else:
test_dir = r'd:\tmp\bowl\results_test\dpn_softmax3\merged'
# labels_dir = r'/home/albu/dev/bowl/train_imgs/labels_all6'
labels_dir = r'D:\dsbowl\train_imgs\labels_all6'
# vgg = r'D:\tmp\bowl\vgg2folds\predict'
new_test_ids = []
rles = []
im_names = os.listdir(test_dir)
# im_names = [im for im in im_names if not im.startswith('jw-')]
test_ids = [os.path.splitext(i)[0] for i in im_names]
preds_test = [imread(os.path.join(test_dir, im), mode='RGB') for im in im_names]
# vgg_data = [imread(os.path.join(vgg, im), mode='RGB') for im in im_names]
# dts = [imread(os.path.join(dt_dir, im), mode='L') for im in im_names]
if oof:
pred_labels = [cv2.imread(os.path.join(labels_dir, os.path.splitext(im)[0] + '.tif'), cv2.IMREAD_UNCHANGED) for im in im_names]
scores = []
for n, id_ in enumerate(test_ids):
# cv2.imshow('b', preds_test[n][...,2])
# cv2.imshow('r', preds_test[n][...,0])
# cv2.waitKey()
test_img = wsh(preds_test[n][...,2] / 255., 0.3, 1 - preds_test[n][...,1] / 255., preds_test[n][...,2] / 255)
# test_img = postprocess_victor(preds_test[n])
if oof:
test_img = ndi.label(test_img, output=np.uint32)[0]
score = calc_score(pred_labels[n], test_img)[0]
scores.append(score)
else:
cv2.imwrite(os.path.join(r'D:\tmp\bowl\res_bin', im_names[n]), (test_img > 0).astype(np.uint8) * 255)
rle = list(prob_to_rles(test_img))
rles.extend(rle)
new_test_ids.extend([id_] * len(rle))
if oof:
print(np.mean(scores))
else:
sub = pd.DataFrame()
sub['ImageId'] = new_test_ids
sub['EncodedPixels'] = pd.Series(rles).apply(lambda x: ' '.join(str(y) for y in x))
sub.to_csv('sub-dsbowl2018-1.csv', index=False)
Datasets
Models
