Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,101 @@
|
|
1 |
---
|
2 |
license: cc-by-4.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: cc-by-4.0
|
3 |
+
task_categories:
|
4 |
+
- question-answering
|
5 |
+
tags:
|
6 |
+
- 3D vision
|
7 |
+
- embodied AI
|
8 |
+
size_categories:
|
9 |
+
- 10K<n<100K
|
10 |
---
|
11 |
+
|
12 |
+
SQA3D: Situated Question Answering in 3D Scenes
|
13 |
+
===
|
14 |
+
1. Download the [SQA3D dataset](https://zenodo.org/record/7544818/files/sqa_task.zip?download=1) under `assets/data/`. The following files should be used:
|
15 |
+
```plain
|
16 |
+
./assets/data/sqa_task/balanced/*
|
17 |
+
./assets/data/sqa_task/answer_dict.json
|
18 |
+
```
|
19 |
+
|
20 |
+
2. The dataset has been splited into `train`, `val` and `test`. For each category, we offer both question file, ex. `v1_balanced_questions_train_scannetv2.json`, and annotations, ex. `v1_balanced_sqa_annotations_train_scannetv2.json`
|
21 |
+
|
22 |
+
- The format of question file:
|
23 |
+
|
24 |
+
Run the following code:
|
25 |
+
```python
|
26 |
+
import json
|
27 |
+
q = json.load(open('v1_balanced_questions_train_scannetv2.json', 'r'))
|
28 |
+
# Print the total number of questions
|
29 |
+
print('#questions: ', len(q['questions']))
|
30 |
+
print(q['questions'][0])
|
31 |
+
```
|
32 |
+
The output is:
|
33 |
+
```json
|
34 |
+
{
|
35 |
+
"alternative_situation":
|
36 |
+
[
|
37 |
+
"I stand looking out of the window in thought and a radiator is right in front of me.",
|
38 |
+
"I am looking outside through the window behind the desk."
|
39 |
+
],
|
40 |
+
"question": "What color is the desk to my right?",
|
41 |
+
"question_id": 220602000000,
|
42 |
+
"scene_id": "scene0380_00",
|
43 |
+
"situation": "I am facing a window and there is a desk on my right and a chair behind me."
|
44 |
+
}
|
45 |
+
```
|
46 |
+
The following fileds are **useful**: `question`, `question_id`, `scene_id`, `situation`.
|
47 |
+
|
48 |
+
- The format of annotations:
|
49 |
+
|
50 |
+
Run the following code:
|
51 |
+
```python
|
52 |
+
import json
|
53 |
+
a = json.load(open('v1_balanced_sqa_annotations_train_scannetv2.json', 'r'))
|
54 |
+
# Print the total number of annotations, should be the same as questions
|
55 |
+
print('#annotations: ', len(a['annotations']))
|
56 |
+
print(a['annotations'][0])
|
57 |
+
```
|
58 |
+
The output is
|
59 |
+
```json
|
60 |
+
{
|
61 |
+
"answer_type": "other",
|
62 |
+
"answers":
|
63 |
+
[
|
64 |
+
{
|
65 |
+
"answer": "brown",
|
66 |
+
"answer_confidence": "yes",
|
67 |
+
"answer_id": 1
|
68 |
+
}
|
69 |
+
],
|
70 |
+
"position":
|
71 |
+
{
|
72 |
+
"x": -0.9651003385573296,
|
73 |
+
"y": -1.2417634435553606,
|
74 |
+
"z": 0
|
75 |
+
},
|
76 |
+
"question_id": 220602000000,
|
77 |
+
"question_type": "N/A",
|
78 |
+
"rotation":
|
79 |
+
{
|
80 |
+
"_w": 0.9950041652780182,
|
81 |
+
"_x": 0,
|
82 |
+
"_y": 0,
|
83 |
+
"_z": 0.09983341664682724
|
84 |
+
},
|
85 |
+
"scene_id": "scene0380_00"
|
86 |
+
}
|
87 |
+
```
|
88 |
+
The following fields are **useful**: `answers[0]['answer']`, `question_id`, `scene_id`.
|
89 |
+
|
90 |
+
**Note**: To find the answer of a question in the question file, you need to use lookup with `question_id`.
|
91 |
+
|
92 |
+
3. We provide the mapping between answers and class labels in `answer_dict.json`
|
93 |
+
```python
|
94 |
+
import json
|
95 |
+
j = json.load(open('answer_dict.json', 'r'))
|
96 |
+
print('Total classes: ', len(j[0]))
|
97 |
+
print('The class label of answer \'table\' is: ', j[0]['table'])
|
98 |
+
print('The corresponding answer of class 123 is: ', j[1]['123'])
|
99 |
+
```
|
100 |
+
|
101 |
+
4. Loader, model and training code can be found at https://github.com/SilongYong/SQA3D
|