print 'Invalid NeuralTalk results file'
sys.exit(-1)
captions_dict = pickle.load(open(captions_dict_path))
generated_captions_list = open(nt_results_path, 'r').readlines()
tp = 0
for gc_item in generated_captions_list:
frame_filename, generated_caption = gc_item.split(SEPARATOR)
frame_filename = frame_filename.strip()
generated_caption = generated_caption.strip()
video_filename = get_videoname(frame_filename)
if not video_filename in captions_dict:
print 'Video was not found in dictionary:\n%s' % (video_filename)
continue
original_caption = captions_dict[video_filename]
generated_verbs = nlpa.extract_lemmatized_verbs(generated_caption)
original_caption = nlpa.remove_invalid_unicode(
original_caption) #clean-up caption
original_verbs = set(nlpa.extract_lemmatized_verbs(original_caption))
for gverb in generated_verbs:
if gverb in original_verbs:
tp += 1
break
tp_rate = tp / len(generated_captions_list)
print 'TP: %d' % tp
print 'Total captions: %d' % len(generated_captions_list)
print 'True positive rate: %04f' % (tp_rate)
if not os.path.isfile(nt_results_path):
print 'Invalid NeuralTalk results file'
sys.exit(-1)
captions_dict = pickle.load(open(captions_dict_path))
generated_captions_list = open(nt_results_path, 'r').readlines()
tp = 0
for gc_item in generated_captions_list:
frame_filename, generated_caption = gc_item.split(SEPARATOR)
frame_filename = frame_filename.strip()
generated_caption = generated_caption.strip()
video_filename = get_videoname(frame_filename)
if not video_filename in captions_dict:
print 'Video was not found in dictionary:\n%s'%(video_filename)
continue
original_caption = captions_dict[video_filename]
generated_verbs = nlpa.extract_lemmatized_verbs(generated_caption)
original_caption = nlpa.remove_invalid_unicode(original_caption) #clean-up caption
original_verbs = set(nlpa.extract_lemmatized_verbs(original_caption))
for gverb in generated_verbs:
if gverb in original_verbs:
tp += 1
break
tp_rate = tp/len(generated_captions_list)
print 'TP: %d'%tp
print 'Total captions: %d'%len(generated_captions_list)
print 'True positive rate: %04f'%(tp_rate)
all_srt_path = '/Users/zal/CMU/Fall2015/HCMMML/FinalProject/Dataset/MontrealVideoAnnotationDataset/M-VAD/srt_files/all_srt'
dict_filename = '/Users/zal/CMU/Fall2015/HCMMML/FinalProject/Repository/DataProcessing/all_captions_dict.p'
inv_dict_filename = '/Users/zal/CMU/Fall2015/HCMMML/FinalProject/Repository/DataProcessing/all_captions_inv_dict.p'
verbs_dict_filename = '/Users/zal/CMU/Fall2015/HCMMML/FinalProject/Repository/DataProcessing/all_captions_verbs_dict.p'
captions_dict = {}
captions_inv_dict = {}
verbs_dict = {}
for srt_file in glob.glob(join(all_srt_path,'*.srt')):
movie_name = os.path.splitext(os.path.basename(srt_file))[0]
all_lines = open(join(all_srt_path, srt_file),'r').readlines()
for i in range(0,len(all_lines),4):
file_name = all_lines[i].strip()
caption = all_lines[i+2].strip()
#Captions dict Key:VideoFileName Value:Caption
captions_dict[file_name]=caption
#Captions Inv Index Key:Caption Value:Video FileName
captions_inv_dict[caption]=file_name
#Inverted dict based on verbs Key: lemmatizedVerb Value: [(filename, caption), ...]
caption_verbs = nlpa.extract_lemmatized_verbs(caption.decode('utf8'))
for verb in caption_verbs:
if verb not in verbs_dict:
verbs_dict[verb] = [(file_name, caption)]
else:
verbs_dict[verb].append((file_name, caption))
pickle.dump(captions_dict,open(dict_filename, 'wb'))
pickle.dump(captions_inv_dict,open(inv_dict_filename, 'wb'))
pickle.dump(verbs_dict, open(verbs_dict_filename, 'wb'))