jdsannchao
commited on
Commit
•
7915082
1
Parent(s):
7173aa8
Update README.md
Browse files
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
|
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 |
-
|
|
|
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 |
-
|
|
|
|
|
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. }
|