def detection_error_example():
big_crat = np.asarray(tiles['3_25'].crop((800, 725, 1450, 1350)))
fig, ax = plt.subplots(1, 3, figsize=(12, 4))
fig.suptitle('Erroneous Detection Criterion')
ax[0].imshow(big_crat, cmap='Greys_r')
ax[0].set_title('Diameter Error')
circle = plt.Circle((322, 329), 144, fill=False, color='green', linewidth=2);
ax[0].add_artist(circle);
circle = plt.Circle((322, 329), 201, fill=False, color='red', linewidth=2);
ax[0].add_artist(circle);
circle = plt.Circle((322, 329), 87, fill=False, color='red', linewidth=2);
ax[0].add_artist(circle);
ax[1].imshow(big_crat, cmap='Greys_r')
ax[1].set_title('Location Error')
circle = plt.Circle((322, 329), 144, fill=False, color='green', linewidth=2);
ax[1].add_artist(circle);
circle = plt.Circle((322, 444), 144, fill=False, color='red', linewidth=2);
ax[1].add_artist(circle);
ax[2].imshow(big_crat, cmap='Greys_r')
ax[2].set_title('Combination of Error Types')
circle = plt.Circle((322, 329), 144, fill=False, color='green', linewidth=2);
ax[2].add_artist(circle);
circle = plt.Circle((256, 395), 111, fill=False, color='red', linewidth=2);
ax[2].add_artist(circle);
circle = plt.Circle((388, 263), 177, fill=False, color='red', linewidth=2);
ax[2].add_artist(circle);
ax[0], ax[1], ax[2] = remove_ticks(ax[0]), remove_ticks(ax[1]), remove_ticks(ax[2])
handles = []
handles.append(mpatches.Patch(color='green', label='Perfect Detection'))
handles.append(mpatches.Patch(color='red', label='Rejected by Criteria'))
plt.legend(handles=handles);
def plot_mask():
fig, ax = plt.subplots(ncols=2, figsize=(6, 3))
fig.suptitle('Sample Image and Target Mask');
ax[0].imshow(samp_img, cmap='Greys_r')
ax[0] = test_human.remove_ticks(ax[0])
ax[1].imshow(samp_mask, cmap='CMRmap')
ax[1] = test_human.remove_ticks(ax[1])
plt.show()
def plot_filters():
fig, ax = plt.subplots(ncols=2, figsize=(9, 5))
fig.suptitle('Automatically-learned Image Filters');
ax[0].imshow(samp_img, cmap='Greys_r')
ax[0] = test_human.remove_ticks(ax[0])
ax[1].imshow(filt_img, cmap='CMRmap')
ax[1] = test_human.remove_ticks(ax[1])
plt.tight_layout();
plt.show()
def inspect_circle_output(id_no):
"""Displays a previously-saved prediction for inspection
by loading it from hard drive.
"""
base_path = './tiles/keras_folders/test'
prediction = np.load('./tiles/predictions/{}.npy'.format(id_no))
source_image = crop_square(Image.open(base_path+'/{}.png'.format(id_no)), 172, orgn=(42, 42))
ground_truth = Image.open(base_path+'_mask/{}_mask.png'.format(id_no))
circles = FastCircles()
craters = circles(prediction)
extraction = build_target(craters)
fig, ax = plt.subplots(ncols=4, figsize=(12, 3))
fig.suptitle('Desirable Behavior from Circle Procedure')
ax = [remove_ticks(x) for x in ax]
ax[0].imshow(np.array(source_image), cmap='Greys_r');
ax[0].set_title('Input image');
ax[1].imshow(np.array(ground_truth), cmap='CMRmap');
ax[1].set_title('"Ground Truth"');
ax[2].imshow(prediction, cmap='CMRmap');
ax[2].set_title('Model Prediction');
ax[3].imshow(extraction, cmap='CMRmap');
ax[3].set_title('Extracted Circles');
plt.show()
return
def plot_tiles(tiles, tile_names, regions):
"""Plots the tiles for demonstration purposes."""
fig, axes = plt.subplots(nrows=2, ncols=3, figsize=(9,6))
axes = axes.reshape(6,)
for i, ax in enumerate(axes):
ax = test_human.remove_ticks(ax)
tile = tile_names[i][4:]
ax.set_title(tile + ', ' + regions[i])
img = tiles[tile]
ax.imshow(np.array(img), cmap='Greys_r')
def display_proposals(proposals=tp, title='title', num_imgs=5):
fig, ax = plt.subplots(1, num_imgs, figsize=(num_imgs, 2));
fig.suptitle(title);
num = 0
for axis in ax:
img = test_human.get_image(human_performance[proposals]['id'].iloc[num])
axis.imshow(img, cmap='Greys')
axis = test_human.remove_ticks(axis)
plt.tight_layout()
num += 1
def show_dbscan():
pixels = np.argwhere(blob_image)
#indices = [i for i in range(len(pixels))]
#idx = np.random.choice(indices, round(len(pixels)/10))
#pixels = pixels[idx]
dbscan = DBSCAN(eps=1)
pred = dbscan.fit_predict(pixels)
cs = ['r', 'b', 'green', 'yellow', 'cyan', 'violet', 'black', 'orange']
cs = cs * 100
print('number of clusters: {}'.format(len(np.unique(pred))))
fig, ax = plt.subplots()
for cluster in np.unique(pred):
indices = np.argwhere(np.where(pred == cluster, 1, 0));
ax.scatter(x=pixels[indices, 1], y = (-pixels[indices,0]+150), color=cs[cluster]);
ax = remove_ticks(ax)
plt.show()
def overlay_mask(id_no, train=True):
"""Fetches image and accompanying mask and overlays them. Confirms mask
is properly generated.
"""
path = './tiles/keras_folders/'
if train:
path += 'train/'
else:
path += 'test/'
img = np.array(Image.open(path + '{}.png'.format(id_no)))
mask = np.array(Image.open(path[:-1] + '_mask/{}_mask.png'.format(id_no)))
mask = pad_mask(mask)
fig, ax = plt.subplots()
ax.imshow(img, cmap='Greys', alpha=1)
ax.imshow(mask, cmap='viridis', alpha=.15)
ax = remove_ticks(ax)
ax.set_title('Target "ground truth" overlaid on input image')
def inspect_detection(id_no):
"""Displays a previously-saved prediction for inspection
by loading it from hard drive.
"""
base_path = './tiles/keras_folders/test'
prediction = np.load('./tiles/predictions/{}.npy'.format(id_no))
source_image = crop_square(Image.open(base_path+'/{}.png'.format(id_no)), 172, orgn=(42, 42))
ground_truth = Image.open(base_path+'_mask/{}_mask.png'.format(id_no))
fig, ax = plt.subplots(ncols=3, figsize=(8, 4))
fig.suptitle('Results for Tile {}'.format(id_no))
ax = [remove_ticks(x) for x in ax]
ax[0].imshow(np.array(source_image), cmap='Greys_r');
ax[0].set_title('Input image');
ax[1].imshow(np.array(ground_truth), cmap='CMRmap');
ax[1].set_title('"Ground Truth"');
ax[2].imshow(prediction, cmap='CMRmap');
ax[2].set_title('Model Prediction');
plt.tight_layout()
plt.show()
return
rr, cc = circle(y, x, r)
try:
image[rr, cc] = 255
except:
pass
return image
if __name__ == '__main__':
path = './tiles/mask/'
for tile in tiles:
mask = build_target(all_craters[all_craters.tile == tile])
fig, ax = plt.subplots(ncols=2, figsize=(10, 5))
ax[0].set_title(tile)
ax[0].imshow(mask, cmap='Greys')
ax[0] = remove_ticks(ax[0])
ax[1].set_title(tile)
ax[1].imshow(np.where(mask == 255, tiles[tile], 0), cmap='Greys')
ax[1] = remove_ticks(ax[1])
image = Image.fromarray(mask)
image.save(path + '{}_mask.bmp'.format(tile))
# In[3]:
tf_dir = './tiles/t_tiles/'
# In[4]:
def random_id(K=5):
"""Returns a string of random characters (numeric and upper alpha)