def plot_train(train_path='../data/stage1_train/'):
"""
Plot and save true boundaries of nuclei for each training image
"""
if not os.path.isdir('../train_images'):
os.mkdir('../train_images')
df = pd.read_csv('../data/train_df.csv')
df = df.set_index('img_id')
train_ids = os.listdir(train_path)
dataset_train = KaggleDataset()
dataset_train.load_shapes(train_ids, train_path)
dataset_train.prepare()
image_ids = dataset_train.image_ids
for mm, image_id in tqdm(enumerate(image_ids)):
image = dataset_train.load_image(image_id, color='RGB')
masks, class_ids = dataset_train.load_mask(image_id)
fig = plot_boundary(image, true_masks=masks)
fig.savefig('../train_images/train_' + str(mm) + '.png',
bbox_inches='tight')
plt.close()
def pred_n_plot_test(model,
config,
test_path='../data/stage2_test_final/',
save_plots=False):
"""
Predicts nuclei for each image, draws the boundaries and saves in images folder.
"""
# Create images folder if doesn't exist.
if not os.path.isdir('../images'):
os.mkdir('../images')
# Load test dataset
test_ids = os.listdir(test_path)
dataset_test = KaggleDataset()
dataset_test.load_shapes(test_ids, test_path)
dataset_test.prepare()
new_test_ids = []
rles = []
id = 0
# No masks prediction counter
no_masks = 0
for mm, image_id in tqdm(enumerate((dataset_test.image_ids))):
# Load the image
print("image_id ", image_id)
image = dataset_test.load_image(image_id, color=config.IMAGE_COLOR)
# Image name for submission rows.
image_id = dataset_test.image_info[image_id]['img_name']
height, width = image.shape[:2]
print(image_id)
# result = ensemble_prediction(model, config, image)
result = cluster_prediction(model, config, image)
# result = model.detect([image], verbose=0, mask_threshold=config.DETECTION_MASK_THRESHOLD)[0]
# Clean overlaps and apply some post-processing
result = postprocess_masks(result, image)
# If there is no masks then try to predict on scaled image
if result['masks'].sum() < 2:
H, W = image.shape[:2]
scaled_img = np.zeros([4 * H, 4 * W, 3], np.uint8)
scaled_img[:H, :W] = image
result = cluster_prediction(model, config, scaled_img)
result['masks'] = result['masks'][:H, :W]
result = postprocess_masks(result, image)
if result['masks'].sum() < 1:
no_masks += 1
# RLE encoding
rle = list(prob_to_rles(result['masks'], height, width))
new_test_ids.extend([image_id] * len(rle))
rles.extend(rle)
# if (mm+1) % 100 == 0:
# print('Number of completed images', mm+1)
print(type(result['masks']))
print("shape_of_something ", result['masks'].shape)
print("type_of_image ", type(image))
print("image_shape ", image.shape)
print("image_id ", image_id)
print(image_id.split('_')[0])
if save_plots:
fig = plt.figure()
gs = gridspec.GridSpec(1, 1)
plot_boundary(image,
true_masks=None,
pred_masks=result['masks'],
ax=fig.add_subplot(gs[0]))
fig.savefig('../images/' + str(image_id) + '.png',
bbox_inches='tight')
# plotting and saving the boundaries
fig2 = plt.figure()
gs = gridspec.GridSpec(1, 1)
msk = np.zeros(image.shape)
id = plot_boundary_mask(msk,
true_masks=None,
pred_masks=result['masks'],
ax=fig2.add_subplot(gs[0]),
image_id=image_id,
id=id)
fig2.savefig('../masks2/' + str(image_id) + '.png',
bbox_inches='tight')
plt.close()
sub = pd.DataFrame()
sub['ImageId'] = new_test_ids
sub['EncodedPixels'] = pd.Series(rles).apply(
lambda x: ' '.join(str(y) for y in x))
fname = '../data/' + test_path.split('/')[-2] + '_submission.csv'
sub.to_csv(fname, index=False)
print('Number of rows:', len(sub))
print('No mask prediction for', no_masks, 'images')
config.IMAGES_PER_GPU = 1
config.BATCH_SIZE = 1
config.display()
# Predict using the last weights in training directory
model = get_model(config)
# Predict using pre-trained weights
# model = get_model(config, model_path='../data/kaggle_bowl.h5')
# Ininitialize validation dataset
train_path = '/home/videsh/Downloads/Chandan/data_science_bowl_data/stage1_train/'
train_list, val_list = train_validation_split(train_path,
seed=11,
test_size=0.1)
dataset_val = KaggleDataset()
dataset_val.load_shapes(val_list, train_path)
dataset_val.prepare()
# initialize stage 1 testing dataset
dataset_val = KaggleDataset()
val_path = '/home/videsh/Downloads/Chandan/data_science_bowl_data/stage1_test/'
val_list = os.listdir(val_path)
dataset_val.load_shapes(val_list, val_path)
dataset_val.prepare()
# Evaluate the model performance and plot boundaries for the predictions
# eval_n_plot_val(model, config, dataset_val, save_plots=True)
# Predict and plot boundaries for stage1 test
# ls = os.listdir('/home/videsh/Downloads/Chandan/data science bowl data/small_test_data/')
# for f in ls: