Update README.md
Browse files
README.md
CHANGED
@@ -9,9 +9,13 @@ language:
|
|
9 |
|
10 |
TiO is an Interactive Visual Grounding Model for Disambiguation. (WIP)
|
11 |
|
12 |
-
## Online /
|
|
|
|
|
|
|
13 |
|
14 |
```python
|
|
|
15 |
from transformers import AutoModel, AutoTokenizer, AutoImageProcessor
|
16 |
|
17 |
model_id = "jxu124/TiO"
|
@@ -25,25 +29,32 @@ model = AutoModel.from_pretrained(
|
|
25 |
)
|
26 |
tokenizer = AutoTokenizer.from_pretrained(model_id, use_fast=False)
|
27 |
image_processor = AutoImageProcessor.from_pretrained(model_id)
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
```
|
32 |
|
33 |
## Mini-Example
|
34 |
```python
|
|
|
35 |
from transformers import AutoModel, AutoTokenizer, AutoImageProcessor
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
from PIL import Image
|
37 |
from io import BytesIO
|
38 |
import torch
|
39 |
import requests
|
40 |
|
41 |
-
# Load model, tokenizer, image_processor
|
42 |
-
tokenizer = AutoTokenizer.from_pretrained("jxu124/TiO", use_fast=False)
|
43 |
-
image_processor = AutoImageProcessor.from_pretrained("jxu124/TiO")
|
44 |
-
model = AutoModel.from_pretrained("jxu124/TiO", trust_remote_code=True)
|
45 |
-
model = model.to(torch.float16).cuda() # It would be faster.
|
46 |
-
|
47 |
# Prepare example
|
48 |
image = Image.open(BytesIO(requests.get("http://images.cocodataset.org/val2014/COCO_val2014_000000429913.jpg").content))
|
49 |
text = """\
|
@@ -64,25 +75,28 @@ print(tokenizer.batch_decode(gen, skip_special_tokens=True).replace("not yet.",
|
|
64 |
|
65 |
Guesser(grounding):
|
66 |
```python
|
67 |
-
text = """
|
|
|
|
|
68 |
human: look that man in white!
|
69 |
agent: is he the one who just threw the ball?
|
70 |
-
human: yes. I mean the pitcher
|
71 |
-
"""
|
72 |
```
|
73 |
|
74 |
Questioner(question generation):
|
75 |
```python
|
76 |
-
text = """
|
77 |
-
|
78 |
-
|
|
|
79 |
```
|
80 |
|
81 |
Oracle(answering):
|
82 |
```python
|
83 |
-
text = """
|
|
|
|
|
84 |
agent: look that man in white!
|
85 |
-
human: is he the one who just threw the ball?
|
86 |
-
#region: <bin_847> <bin_319> <bin_923> <bin_467>
|
87 |
-
"""
|
88 |
```
|
|
|
9 |
|
10 |
TiO is an Interactive Visual Grounding Model for Disambiguation. (WIP)
|
11 |
|
12 |
+
## Online / Offline Demo
|
13 |
+
|
14 |
+
- [Colab Online Demo](https://colab.research.google.com/drive/195eDITKi6dahnVz8Cum91sNUCF_lFle8?usp=sharing) - Free T4 is available on Google Colab.
|
15 |
+
- Gradio Offline Demo:
|
16 |
|
17 |
```python
|
18 |
+
import os; os.system("pip3 install transformers accelerate bitsandbytes gradio fire")
|
19 |
from transformers import AutoModel, AutoTokenizer, AutoImageProcessor
|
20 |
|
21 |
model_id = "jxu124/TiO"
|
|
|
29 |
)
|
30 |
tokenizer = AutoTokenizer.from_pretrained(model_id, use_fast=False)
|
31 |
image_processor = AutoImageProcessor.from_pretrained(model_id)
|
32 |
+
|
33 |
+
# ---- setup gradio demo ----
|
34 |
+
model.get_gradio_demo(tokenizer, image_processor).queue(max_size=20).launch(server_name="0.0.0.0", server_port=7860)
|
35 |
```
|
36 |
|
37 |
## Mini-Example
|
38 |
```python
|
39 |
+
import os; os.system("pip3 install transformers accelerate bitsandbytes gradio fire")
|
40 |
from transformers import AutoModel, AutoTokenizer, AutoImageProcessor
|
41 |
+
|
42 |
+
model_id = "jxu124/TiO"
|
43 |
+
model = AutoModel.from_pretrained(
|
44 |
+
model_id,
|
45 |
+
trust_remote_code=True,
|
46 |
+
torch_dtype=torch.float16,
|
47 |
+
device_map='cuda'
|
48 |
+
)
|
49 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id, use_fast=False)
|
50 |
+
image_processor = AutoImageProcessor.from_pretrained(model_id)
|
51 |
+
|
52 |
+
# ---- mini example ----
|
53 |
from PIL import Image
|
54 |
from io import BytesIO
|
55 |
import torch
|
56 |
import requests
|
57 |
|
|
|
|
|
|
|
|
|
|
|
|
|
58 |
# Prepare example
|
59 |
image = Image.open(BytesIO(requests.get("http://images.cocodataset.org/val2014/COCO_val2014_000000429913.jpg").content))
|
60 |
text = """\
|
|
|
75 |
|
76 |
Guesser(grounding):
|
77 |
```python
|
78 |
+
text = """\
|
79 |
+
#instruction: which region does the context describe?
|
80 |
+
#context:
|
81 |
human: look that man in white!
|
82 |
agent: is he the one who just threw the ball?
|
83 |
+
human: yes. I mean the pitcher."""
|
|
|
84 |
```
|
85 |
|
86 |
Questioner(question generation):
|
87 |
```python
|
88 |
+
text = """\
|
89 |
+
#instruction: guess what I want?
|
90 |
+
#context:
|
91 |
+
human: look that man in white!"""
|
92 |
```
|
93 |
|
94 |
Oracle(answering):
|
95 |
```python
|
96 |
+
text = """\
|
97 |
+
#instruction: answer the question based on the region.
|
98 |
+
#context:
|
99 |
agent: look that man in white!
|
100 |
+
human: is he the one who just threw the ball?
|
101 |
+
#region: <bin_847> <bin_319> <bin_923> <bin_467>"""
|
|
|
102 |
```
|