Start to add test subset

gqa-lxmert.py

@@ -44,12 +44,14 @@ seeking to address key shortcomings of previous visual question answering (VQA)
 
 _URLS = {
     "train": "https://nlp.cs.unc.edu/data/lxmert_data/gqa/train.json",
-    "dev": "https://nlp.cs.unc.edu/data/lxmert_data/gqa/valid.json",
-    "feat": "https://nlp.cs.unc.edu/data/lxmert_data/vg_gqa_imgfeat/vg_gqa_obj36.zip",
+    "val": "https://nlp.cs.unc.edu/data/lxmert_data/gqa/valid.json",
+    "trainval_feat": "https://nlp.cs.unc.edu/data/lxmert_data/vg_gqa_imgfeat/vg_gqa_obj36.zip",
+    "test": "https://nlp.cs.unc.edu/data/lxmert_data/gqa/testdev.json",
+    "test_feat": "https://nlp.cs.unc.edu/data/lxmert_data/vg_gqa_imgfeat/gqa_testdev_obj36.zip",
     "ans2label": "https://raw.githubusercontent.com/airsplay/lxmert/master/data/gqa/trainval_ans2label.json",
 }
 
-_FEAT_PATH = "vg_gqa_imgfeat/vg_gqa_obj36.tsv"
+_FEAT_PATH = "vg_gqa_imgfeat/gqa_testdev_obj36.tsv"
 
 FIELDNAMES = [
     "img_id", "img_h", "img_w", "objects_id", "objects_conf", "attrs_id", "attrs_conf", "num_boxes", "boxes", "features"
@@ -132,8 +134,15 @@ class GqaLxmert(datasets.GeneratorBasedBuilder):
         with open(filepath, encoding="utf-8") as f:
             gqa = json.load(f)
             for id_, d in enumerate(gqa):
-                img_features = self.id2features[d["img_id"]]
-                label = self.ans2label[next(iter(d["label"]))]
+                try:
+                    img_features = self.id2features[d["img_id"]]
+                except KeyError:
+                    continue
+                label_key = next(iter(d["label"]))
+                if label_key not in self.ans2label:
+                    print ("Ignored one label")
+                    continue
+                label = self.ans2label[label_key]
                 yield id_, {
                     "question": d["sent"],
                     "question_id": d["question_id"],