jdsannchao commited on
Commit
7915082
1 Parent(s): 7173aa8

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +9 -3
README.md CHANGED
@@ -6,21 +6,27 @@ Here we create two datasets (from existing datasets: CLEVRER, VisualGenome) for
6
 
7
  ### CLEVRER
8
 
9
- CLEVRER has QA pairs for each 5000 training video
10
  ```json
11
  {'video_filename': int, 'scene_index': str (same as filename), 'questions': list [{'question_type': , 'question_subtype': , 'question_text': , 'answer_text': , 'program'(question attributes): }]}
12
  ```
13
  We select 'descriptive' type, 'count' subtype questions, they are object counting.
 
14
  CLEVRER contains both positive questions and negative (non-exist) questions, so no need to construct negative samples.
 
15
  some questions are 'event' specific, counting moving/stationary objects when a certain event happens. i.e., 'How many objects are stationary when the yellow object enters the scene?'
16
- downloading videos: http://clevrer.csail.mit.edu/
 
17
 
18
 
19
  ### VisualGenome
20
 
21
  We generate some negative questions for non-exist objects in the image. We use the version 1 image sets. Download from: https://homes.cs.washington.edu/~ranjay/visualgenome/api.html
 
22
  VisualGenome has 100K+ images. And for the objects in the image, there are attributes, We only focus on the color attributes.
23
- There are in total 11K+ possible objects. for each image, I add 3 non-exist objects and 1 non-exist attribute for existing objects as negative samples.
 
 
24
  In the original qa dataset, VG has Object Counting questions, we also include them here, with the 'orig_qa'=='Yes'. For those negative questions we generated, 'orig_qa' =='No'.
25
  ```json
26
  {'img_id': str, 'orig_qa': Yes/No, 'question_text': 'How many <attribute> <object in plural form> are there? ', 'answer_text': Numbers. or None. }
 
6
 
7
  ### CLEVRER
8
 
9
+ CLEVRER has QA pairs for each 5000 training videos.
10
  ```json
11
  {'video_filename': int, 'scene_index': str (same as filename), 'questions': list [{'question_type': , 'question_subtype': , 'question_text': , 'answer_text': , 'program'(question attributes): }]}
12
  ```
13
  We select 'descriptive' type, 'count' subtype questions, they are object counting.
14
+
15
  CLEVRER contains both positive questions and negative (non-exist) questions, so no need to construct negative samples.
16
+
17
  some questions are 'event' specific, counting moving/stationary objects when a certain event happens. i.e., 'How many objects are stationary when the yellow object enters the scene?'
18
+
19
+ downloading videos from: http://clevrer.csail.mit.edu/
20
 
21
 
22
  ### VisualGenome
23
 
24
  We generate some negative questions for non-exist objects in the image. We use the version 1 image sets. Download from: https://homes.cs.washington.edu/~ranjay/visualgenome/api.html
25
+
26
  VisualGenome has 100K+ images. And for the objects in the image, there are attributes, We only focus on the color attributes.
27
+
28
+ There are in total 11K+ possible objects. for each image, I add 3 non-exist objects and 1 non-exist attribute for existing objects as negative samples.
29
+
30
  In the original qa dataset, VG has Object Counting questions, we also include them here, with the 'orig_qa'=='Yes'. For those negative questions we generated, 'orig_qa' =='No'.
31
  ```json
32
  {'img_id': str, 'orig_qa': Yes/No, 'question_text': 'How many <attribute> <object in plural form> are there? ', 'answer_text': Numbers. or None. }