GQA problem

configs/default/components/problems/image_text_to_class/gqa.yml

@@ -0,0 +1,77 @@
+# This file defines the default values for the GQA problem.
+
+####################################################################
+# 1. CONFIGURATION PARAMETERS that will be LOADED by the component.
+####################################################################
+
+# Folder where problem will store data (LOADED)
+data_folder: '~/data/gqa'
+
+# Defines the set (split) that will be used (LOADED)
+# Options: training_0 | training | validation | test_dev | test | challenge | submission (?)
+# Note: test_dev should be used for validation.
+split: training_0
+
+# Flag indicating whether the problem will load and return images (LOADED)
+stream_images: True
+
+# Resize parameter (LOADED)
+# When present, resizes the images from original size to [height, width]
+# Depth remains set to 3.
+resize_image: [224, 224]
+
+# Select applied image preprocessing/augmentations (LOADED)
+# Use one (or more) of the affine transformations:
+# none | normalize | all
+# Accepted formats: a,b,c or [a,b,c]
+image_preprocessing: none
+
+streams:
+  ####################################################################
+  # 2. Keymappings associated with INPUT and OUTPUT streams.
+  ####################################################################
+
+  # Stream containing batch of indices (OUTPUT)
+  # Every problem MUST return that stream.
+  indices: indices
+
+  # Stream containing batch of sample (original) identifiers (OUTPUT)
+  sample_ids: sample_ids
+
+  # Stream containing batch of images (OUTPUT)
+  images: images
+
+  # Stream containing batch of image names (OUTPUT)
+  image_ids: image_ids
+
+  # Stream containing batch of questions (OUTPUT)
+  questions: questions
+
+  # Stream containing targets answers (labels) (OUTPUT)
+  answers: answers
+
+  # Stream containing targets answers consisting of many words (OUTPUT)
+  full_answers: full_answers
+
+  # Stream containing scene descriptions (OUTPUT)
+  #answers: scene_graphs
+
+globals:
+  ####################################################################
+  # 3. Keymappings of variables that will be RETRIEVED from GLOBALS.
+  ####################################################################
+
+  ####################################################################
+  # 4. Keymappings associated with GLOBAL variables that will be SET.
+  ####################################################################
+
+  # Width of the image (SET)
+  input_width: image_width
+  # Height of the image (SET)
+  input_height: image_height
+  # Depth of the image (SET)
+  input_depth: image_depth
+
+  ####################################################################
+  # 5. Keymappings associated with statistics that will be ADDED.
+  ####################################################################

ptp/components/problems/image_text_to_class/clevr.py

@@ -71,10 +71,6 @@ def __init__(self, name, config):
         # Call constructors of parent classes.
         Problem.__init__(self, name, CLEVR, config)
 
-        # (Eventually) download required packages.
-        #nltk.download('punkt')
-        #nltk.download('stopwords')
-
         # Get key mappings of all output streams.
         self.key_images = self.stream_keys["images"]
         self.key_image_ids = self.stream_keys["image_ids"]
@@ -246,7 +242,7 @@ def output_data_definitions(self):
         d[self.key_questions] = DataDefinition([-1, 1], [list, str], "Batch of questions, each being a string consisting of many words [BATCH_SIZE] x [STRING]")
 
         # Add stream with answers.
-        d[self.key_answers]= DataDefinition([-1, 1], [list, str], "Batch of target answers, each being a string consisting of many words [BATCH_SIZE] x [STRING]")
+        d[self.key_answers]= DataDefinition([-1, 1], [list, str], "Batch of target answers, each being a string consisting of sinlge word (label) [BATCH_SIZE] x [STRING]")
 
         return d
 
@@ -271,7 +267,7 @@ def load_dataset(self, source_data_file):
         dataset = []
 
         with open(source_data_file) as f:
-            self.logger.info('Loading samples from {} ...'.format(source_data_file))
+            self.logger.info("Loading samples from '{}'...".format(source_data_file))
             dataset = json.load(f)['questions']
 
         self.logger.info("Loaded dataset consisting of {} samples".format(len(dataset)))