count_objects_in_vqa

object_attribute_classifier_cached_features/count_objects_in_vqa.py

+import ujson
+import nltk
+import os
+
+import constants
+
+
+if __name__=='__main__':
+    with open(constants.vqa_train_anno) as file:
+        vqa_train_anno = ujson.load(file)
+
+    with open(constants.vqa_train_subset_qids) as file:
+        train_subset_qids = ujson.load(file)
+
+    with open(constants.object_labels_json) as file:
+        object_labels = ujson.load(file)
+
+    freq = dict()
+    for label in object_labels.keys():
+        freq[label] = 0
+
+    for qid in train_subset_qids:
+        for noun in vqa_train_anno[qid]['question_nouns']:
+            if noun in freq:
+                freq[noun] += 1
+                break
+
+        for noun in vqa_train_anno[qid]['multiple_choice_answer'].split():
+            if noun in freq:
+                freq[noun] += 1
+                break
+
+
+    freq_filename = os.path.join(
+        '/home/ssd/VisualGenome/restructured',
+        'object_label_freq_in_vqa_train_subset.json')
+
+    with open(freq_filename,'w') as file:
+        ujson.dump(freq, file)
+
+    print freq