def db2patches(output_path, train_label=None, sample_limit=None):
# All human annotations in DB converted to a training set
n_angles = 8
angles = np.linspace(0, 360, num=n_angles, endpoint=False)
extract_size = (256, 256) # wdt, hgt
annotated_samples = db.get_human_annotated_samples(train_label=train_label)
n_annotated_samples = len(annotated_samples)
if sample_limit is not None and n_annotated_samples > sample_limit:
print 'Reducing from %d to sample limit %d...' % (n_annotated_samples,
sample_limit)
annotated_samples = np.random.choice(annotated_samples,
sample_limit,
replace=False)
print 'Extracting patches from %d images to %s...' % (
len(annotated_samples), output_path)
n = 0
for s in tqdm(annotated_samples):
img_filename = os.path.join(config.get_server_image_path(),
s['filename'])
img_name = s['filename'].replace('.', '_')
for annotation in db.get_human_annotations(s['_id']):
allpos = [dbpos2extpos(p) for p in annotation['positions']]
n += extract_positions(img_filename, allpos, output_path, img_name,
angles, extract_size, s['_id'])
print '%d patches extracted.' % n
def annotate(sid):
# Query info from DB
sample_id = ObjectId(sid)
sample_entry = db.get_sample_by_id(sample_id)
dataset_id = sample_entry['dataset_id']
name = sample_entry['name']
readonly = db.is_readonly_dataset_id(dataset_id)
sample_index, sample_count, prev_sample_id, next_sample_id = db.get_sample_index(
dataset_id, sample_id)
# Unknown entry?
if sample_entry is None:
return error_redirect('Unknown entry: "%s".' % str(sample_id))
# Determine data
image_filename = 'images/' + sample_entry['filename']
info_string = ''
# Differential? Then load machine annotations, unless there are human annotations already
annotations = db.get_human_annotations(sample_id)
is_differential = request.args.get('differential')
if is_differential and not annotations:
annotations = db.get_machine_annotations(sample_id)
if len(annotations):
annotations_json = annotations[0]['positions']
# Machine annotations are in a different format
if len(annotations_json) and isinstance(annotations_json[0], list):
annotations_json = [{
'x': a[0],
'y': a[1]
} for a in annotations_json]
else:
annotations_json = []
return render_template("annotate.html",
id=sid,
image_filename=image_filename,
info_string=info_string,
error=pop_last_error(),
height=sample_entry['size'][1],
width=sample_entry['size'][0],
annotations=annotations_json,
margin=96,
dataset_id=str(dataset_id),
sample_index=sample_index,
sample_count=sample_count,
prev_id=str(prev_sample_id),
next_id=str(next_sample_id),
is_differential=is_differential,
readonly=readonly,
name=name)
def export_positions_generator(samples, yield_header=True, human=False):
if yield_header:
yield ','.join(
('sample_name', 'sample_filename', 'center_x_px', 'center_y_px',
'center_x_relative', 'center_y_relatrive')) + '\n'
for s in samples:
sample_name = s.get('name')
sample_filename = s.get('filename')
if human:
annotations = db.get_human_annotations(s['_id'])
else:
annotations = db.get_machine_annotations(s['_id'])
if not annotations:
continue
size = s['size']
for pos in annotations[0]['positions']:
if human:
pos = (pos['x'], pos['y'])
rel_pos = ['%.3f' % (float(p) / s) for p, s in zip(pos, size)]
yield ','.join(
map(str, (sample_name, sample_filename, pos[0], pos[1],
rel_pos[0], rel_pos[1]))) + '\n'
def show_info(sid):
# Query info from DB
sample_id = ObjectId(sid)
sample_entry = db.get_sample_by_id(sample_id)
sample_index, sample_count, prev_sample_id, next_sample_id = db.get_sample_index(sample_entry['dataset_id'],
sample_id)
# Unknown entry?
if sample_entry is None:
return error_redirect('Unknown entry: "%s".' % str(sample_id))
# Allow fixing of machine annotation? Only if not human-annotated and there is a machine annotation with stomata
can_annotate_diff = (not sample_entry.get('annotated')) and (sample_entry.get('machine_position_count'))
# Determine data
refresh = False
filename = sample_entry['filename']
name = sample_entry['name']
dataset_id = sample_entry['dataset_id']
readonly = db.is_readonly_dataset_id(dataset_id)
info_table = []
for info_name, info_key in info_table_entries:
info_value = sample_entry.get(info_key)
if info_value is not None:
info_table.append((info_name, info_value))
annotations = []
if sample_entry['error']:
info_string = Markup('Error: <pre>' + sample_entry['error_string'] + '</pre>')
elif not sample_entry['processed']:
info_string = 'Processing...'
refresh = True
else:
info_string = 'Processed.'
annotation_data = db.get_machine_annotations(sample_id) + db.get_human_annotations(sample_id)
has_image_output = False
for ad in annotation_data:
model_id = ad.get('model_id')
an = {'info_string': ''}
if model_id is not None:
model_data = db.get_model_by_id(model_id)
if model_data is None:
an['info_string'] += '??? Unknown model ???'
an['title'] = 'Unknown'
else:
an['title'] = model_data['name']
an['info_string'] += 'Margin: %d' % (ad.get('margin') or model_data['margin'])
an['image_filename'] = 'heatmaps/' + ad['heatmap_image_filename']
has_image_output = True
else:
an['title'] = 'By user %s' % ad.get('user_name')
if 'scale' in ad:
an['info_string'] += ' - Scale: %.1f' % ad['scale']
positions = ad['positions']
if positions is not None:
an['info_string'] += ' - %d stomata' % len(positions)
annotations += [an]
annotations = list(reversed(annotations))
if not has_image_output:
annotations += [{'image_filename': 'images/' + filename, 'title': 'Input image', 'info_string': ''}]
return render_template("info.html", id=sid, name=name, dataset_id=str(dataset_id), info_string=info_string,
annotations=annotations, error=pop_last_error(), refresh=refresh,
sample_index=sample_index, sample_count=sample_count, prev_id=str(prev_sample_id),
next_id=str(next_sample_id), can_annotate_diff=can_annotate_diff, readonly=readonly,
info_table=info_table)