import glob
import os
import shutil
import subprocess
import sys
import tempfile
from operator import itemgetter
from progressbar import ProgressBar, Percentage, Bar, ETA
from .query import eval_query_with_labels
from .utils import extract_frames
from .utils import generate_index_path
from .utils import read_index_from_path
from .utils import save_index_to_path
def times_to_regions(times, max_total_length=None):
if not times:
return []
regions = []
current_start = times[0]
for t1, t2 in zip(times, times[1:]):
if t2 - t1 > 1:
regions.append((current_start, t1 + 1))
current_start = t2
regions.append((current_start, times[-1] + 1))
if not max_total_length:
return regions
accum = 0
ret = []
for index, (t1, t2) in enumerate(regions):
if accum + (t2-t1) > max_total_length:
ret.append((t1, t1 + max_total_length-accum))
return ret
else:
ret.append((t1, t2))
accum += t2 - t1
return ret
def unique_labels(timed_labels):
ret = set()
for t, labels_list in timed_labels:
ret.update(map(itemgetter(0), labels_list))
return ret
def reverse_index(timed_labels, min_occurrences=2, max_length_per_label=8):
ret = {}
for label in unique_labels(timed_labels):
times = []
for t, labels_list in timed_labels:
raw_labels = map(lambda (l, c): l, labels_list)
if eval_query_with_labels(label, raw_labels):
times.append(t)
if len(times) >= min_occurrences:
ret[label] = times_to_regions(times, max_total_length=max_length_per_label)
return ret
def threshold_labels(timed_labels, min_confidence):
ret = []
for t, label_list in timed_labels:
filtered = filter(lambda (l, c): c > min_confidence, label_list)
if filtered:
ret.append((t, filtered))
return ret
def filter_out_labels(timed_labels, ignore_list):
ret = []
for t, label_list in timed_labels:
filtered = filter(lambda (l, c): l not in ignore_list, label_list)
if filtered:
ret.append((t, filtered))
return ret
def label_video(filename, classifier, sample_rate=1, recreate_index=False):
index_filename = generate_index_path(filename, classifier.model)
if os.path.exists(index_filename) and not recreate_index:
return read_index_from_path(index_filename)
temp_frame_dir, frames = extract_frames(filename, sample_rate=sample_rate)
timed_labels = []
widgets=["Labeling {}: ".format(filename), Percentage(), ' ', Bar(), ' ', ETA()]
pbar = ProgressBar(widgets=widgets, maxval=len(frames)).start()
for index, frame in enumerate(frames):
pbar.update(index)
labels = classifier.classify_image(frame)
if not len(labels):
continue
t = (1./sample_rate) * index
timed_labels.append((t, labels))
shutil.rmtree(temp_frame_dir)
save_index_to_path(index_filename, timed_labels)
return timed_labels
def label_videos(filenames, classifier, sample_rate=1, recreate_index=False):
ret = {}
for filename in filenames:
ret[filename] = label_video(
filename,
classifier,
sample_rate=sample_rate,
recreate_index=recreate_index
)
return ret
def search_labels(timed_labels, query, classifier, sample_rate=1, recreate_index=False, min_confidence=0.3):
timed_labels = threshold_labels(timed_labels, min_confidence)
times = []
for t, labels_list in timed_labels:
raw_labels = map(lambda (l, c): l, labels_list)
if eval_query_with_labels(query, raw_labels):
times.append(t)
return times_to_regions(times)
def search_videos(labels_by_filename, query, classifier, sample_rate=1, recreate_index=False, min_confidence=0.3):
ret = []
for filename, timed_labels in labels_by_filename.items():
ret += map(lambda r: (filename, r), search_labels(
timed_labels,
query,
classifier,
sample_rate=sample_rate,
recreate_index=recreate_index,
min_confidence=min_confidence
))
return ret
