Spaces:
Runtime error
Runtime error
gchhablani
commited on
Commit
•
690384a
1
Parent(s):
0e71038
Fix style
Browse files- app.py +63 -35
- requirements.txt +2 -1
- translate_answer_mapping.py +9 -3
- utils.py +16 -5
app.py
CHANGED
@@ -5,8 +5,17 @@ import json
|
|
5 |
import os
|
6 |
import numpy as np
|
7 |
from streamlit.elements import markdown
|
8 |
-
|
9 |
-
from
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
import matplotlib.pyplot as plt
|
11 |
from mtranslate import translate
|
12 |
from PIL import Image
|
@@ -16,23 +25,30 @@ from session import _get_state
|
|
16 |
|
17 |
state = _get_state()
|
18 |
|
|
|
19 |
@st.cache(persist=True)
|
20 |
def load_model(ckpt):
|
21 |
return FlaxCLIPVisionBertForSequenceClassification.from_pretrained(ckpt)
|
22 |
|
|
|
23 |
@st.cache(persist=True)
|
24 |
def predict(transformed_image, question_inputs):
|
25 |
-
return np.array(model(pixel_values
|
26 |
-
|
|
|
27 |
def softmax(logits):
|
28 |
-
return np.exp(logits)/np.sum(np.exp(logits), axis=0)
|
|
|
29 |
|
30 |
def read_markdown(path, parent="./sections/"):
|
31 |
-
with open(os.path.join(parent,path)) as f:
|
32 |
return f.read()
|
33 |
|
34 |
-
|
35 |
-
|
|
|
|
|
|
|
36 |
code_to_name = {
|
37 |
"en": "English",
|
38 |
"fr": "French",
|
@@ -40,7 +56,7 @@ code_to_name = {
|
|
40 |
"es": "Spanish",
|
41 |
}
|
42 |
|
43 |
-
with open(
|
44 |
answer_reverse_mapping = json.load(f)
|
45 |
|
46 |
|
@@ -52,7 +68,9 @@ st.set_page_config(
|
|
52 |
)
|
53 |
|
54 |
st.title("Multilingual Visual Question Answering")
|
55 |
-
st.write(
|
|
|
|
|
56 |
|
57 |
with st.beta_expander("Usage"):
|
58 |
st.markdown(read_markdown("usage.md"))
|
@@ -60,67 +78,77 @@ with st.beta_expander("Usage"):
|
|
60 |
first_index = 20
|
61 |
# Init Session State
|
62 |
if state.image_file is None:
|
63 |
-
state.image_file = dummy_data.loc[first_index,
|
64 |
-
state.question = dummy_data.loc[first_index,
|
65 |
-
state.answer_label = dummy_data.loc[first_index,
|
66 |
-
state.question_lang_id = dummy_data.loc[first_index,
|
67 |
-
state.answer_lang_id = dummy_data.loc[first_index,
|
68 |
-
|
69 |
-
image_path = os.path.join(
|
70 |
image = plt.imread(image_path)
|
71 |
state.image = image
|
72 |
|
73 |
-
col1, col2 = st.beta_columns([6,4])
|
74 |
|
75 |
-
if col2.button(
|
76 |
sample = dummy_data.sample(1).reset_index()
|
77 |
-
state.image_file = sample.loc[0,
|
78 |
-
state.question = sample.loc[0,
|
79 |
-
state.answer_label = sample.loc[0,
|
80 |
-
state.question_lang_id = sample.loc[0,
|
81 |
-
state.answer_lang_id = sample.loc[0,
|
82 |
|
83 |
-
image_path = os.path.join(
|
84 |
image = plt.imread(image_path)
|
85 |
state.image = image
|
86 |
|
87 |
col2.write("OR")
|
88 |
|
89 |
-
uploaded_file = col2.file_uploader(
|
90 |
if uploaded_file is not None:
|
91 |
-
state.image_file = os.path.join(
|
92 |
state.image = np.array(Image.open(uploaded_file))
|
93 |
|
94 |
|
|
|
95 |
transformed_image = get_transformed_image(state.image)
|
96 |
|
97 |
# Display Image
|
98 |
-
col1.image(state.image, use_column_width=
|
99 |
|
100 |
# Display Question
|
101 |
question = col2.text_input(label="Question", value=state.question)
|
102 |
-
col2.markdown(
|
|
|
|
|
103 |
question_inputs = get_text_attributes(question)
|
104 |
|
105 |
# Select Language
|
106 |
-
options = [
|
107 |
-
state.answer_lang_id = col2.selectbox(
|
|
|
|
|
|
|
|
|
|
|
108 |
# Display Top-5 Predictions
|
109 |
-
with st.spinner(
|
110 |
model = load_model(checkpoints[0])
|
111 |
-
with st.spinner(
|
112 |
logits = predict(transformed_image, dict(question_inputs))
|
113 |
logits = softmax(logits)
|
114 |
labels, values = get_top_5_predictions(logits, answer_reverse_mapping)
|
115 |
translated_labels = translate_labels(labels, state.answer_lang_id)
|
116 |
fig = plotly_express_horizontal_bar_plot(values, translated_labels)
|
117 |
-
st.plotly_chart(fig, use_container_width
|
118 |
|
119 |
|
120 |
st.write(read_markdown("abstract.md"))
|
121 |
st.write(read_markdown("caveats.md"))
|
122 |
st.write("# Methodology")
|
123 |
-
st.image(
|
|
|
|
|
124 |
st.markdown(read_markdown("pretraining.md"))
|
125 |
st.markdown(read_markdown("finetuning.md"))
|
126 |
st.write(read_markdown("challenges.md"))
|
|
|
5 |
import os
|
6 |
import numpy as np
|
7 |
from streamlit.elements import markdown
|
8 |
+
import cv2
|
9 |
+
from model.flax_clip_vision_bert.modeling_clip_vision_bert import (
|
10 |
+
FlaxCLIPVisionBertForSequenceClassification,
|
11 |
+
)
|
12 |
+
from utils import (
|
13 |
+
get_transformed_image,
|
14 |
+
get_text_attributes,
|
15 |
+
get_top_5_predictions,
|
16 |
+
plotly_express_horizontal_bar_plot,
|
17 |
+
translate_labels,
|
18 |
+
)
|
19 |
import matplotlib.pyplot as plt
|
20 |
from mtranslate import translate
|
21 |
from PIL import Image
|
|
|
25 |
|
26 |
state = _get_state()
|
27 |
|
28 |
+
|
29 |
@st.cache(persist=True)
|
30 |
def load_model(ckpt):
|
31 |
return FlaxCLIPVisionBertForSequenceClassification.from_pretrained(ckpt)
|
32 |
|
33 |
+
|
34 |
@st.cache(persist=True)
|
35 |
def predict(transformed_image, question_inputs):
|
36 |
+
return np.array(model(pixel_values=transformed_image, **question_inputs)[0][0])
|
37 |
+
|
38 |
+
|
39 |
def softmax(logits):
|
40 |
+
return np.exp(logits) / np.sum(np.exp(logits), axis=0)
|
41 |
+
|
42 |
|
43 |
def read_markdown(path, parent="./sections/"):
|
44 |
+
with open(os.path.join(parent, path)) as f:
|
45 |
return f.read()
|
46 |
|
47 |
+
def resize_height(image, new_height):
|
48 |
+
h, w, c = image.shape
|
49 |
+
|
50 |
+
checkpoints = ["./ckpt/ckpt-60k-5999"] # TODO: Maybe add more checkpoints?
|
51 |
+
dummy_data = pd.read_csv("dummy_vqa_multilingual.tsv", sep="\t")
|
52 |
code_to_name = {
|
53 |
"en": "English",
|
54 |
"fr": "French",
|
|
|
56 |
"es": "Spanish",
|
57 |
}
|
58 |
|
59 |
+
with open("answer_reverse_mapping.json") as f:
|
60 |
answer_reverse_mapping = json.load(f)
|
61 |
|
62 |
|
|
|
68 |
)
|
69 |
|
70 |
st.title("Multilingual Visual Question Answering")
|
71 |
+
st.write(
|
72 |
+
"[Gunjan Chhablani](https://huggingface.co/gchhablani), [Bhavitvya Malik](https://huggingface.co/bhavitvyamalik)"
|
73 |
+
)
|
74 |
|
75 |
with st.beta_expander("Usage"):
|
76 |
st.markdown(read_markdown("usage.md"))
|
|
|
78 |
first_index = 20
|
79 |
# Init Session State
|
80 |
if state.image_file is None:
|
81 |
+
state.image_file = dummy_data.loc[first_index, "image_file"]
|
82 |
+
state.question = dummy_data.loc[first_index, "question"].strip("- ")
|
83 |
+
state.answer_label = dummy_data.loc[first_index, "answer_label"]
|
84 |
+
state.question_lang_id = dummy_data.loc[first_index, "lang_id"]
|
85 |
+
state.answer_lang_id = dummy_data.loc[first_index, "lang_id"]
|
86 |
+
|
87 |
+
image_path = os.path.join("images", state.image_file)
|
88 |
image = plt.imread(image_path)
|
89 |
state.image = image
|
90 |
|
91 |
+
col1, col2 = st.beta_columns([6, 4])
|
92 |
|
93 |
+
if col2.button("Get a random example"):
|
94 |
sample = dummy_data.sample(1).reset_index()
|
95 |
+
state.image_file = sample.loc[0, "image_file"]
|
96 |
+
state.question = sample.loc[0, "question"].strip("- ")
|
97 |
+
state.answer_label = sample.loc[0, "answer_label"]
|
98 |
+
state.question_lang_id = sample.loc[0, "lang_id"]
|
99 |
+
state.answer_lang_id = sample.loc[0, "lang_id"]
|
100 |
|
101 |
+
image_path = os.path.join("images", state.image_file)
|
102 |
image = plt.imread(image_path)
|
103 |
state.image = image
|
104 |
|
105 |
col2.write("OR")
|
106 |
|
107 |
+
uploaded_file = col2.file_uploader("Upload your image", type=["png", "jpg", "jpeg"])
|
108 |
if uploaded_file is not None:
|
109 |
+
state.image_file = os.path.join("images/val2014", uploaded_file.name)
|
110 |
state.image = np.array(Image.open(uploaded_file))
|
111 |
|
112 |
|
113 |
+
state.image =
|
114 |
transformed_image = get_transformed_image(state.image)
|
115 |
|
116 |
# Display Image
|
117 |
+
col1.image(state.image, use_column_width="always")
|
118 |
|
119 |
# Display Question
|
120 |
question = col2.text_input(label="Question", value=state.question)
|
121 |
+
col2.markdown(
|
122 |
+
f"""**English Translation**: {question if state.question_lang_id == "en" else translate(question, 'en')}"""
|
123 |
+
)
|
124 |
question_inputs = get_text_attributes(question)
|
125 |
|
126 |
# Select Language
|
127 |
+
options = ["en", "de", "es", "fr"]
|
128 |
+
state.answer_lang_id = col2.selectbox(
|
129 |
+
"Answer Language",
|
130 |
+
index=options.index(state.answer_lang_id),
|
131 |
+
options=options,
|
132 |
+
format_func=lambda x: code_to_name[x],
|
133 |
+
)
|
134 |
# Display Top-5 Predictions
|
135 |
+
with st.spinner("Loading model..."):
|
136 |
model = load_model(checkpoints[0])
|
137 |
+
with st.spinner("Predicting..."):
|
138 |
logits = predict(transformed_image, dict(question_inputs))
|
139 |
logits = softmax(logits)
|
140 |
labels, values = get_top_5_predictions(logits, answer_reverse_mapping)
|
141 |
translated_labels = translate_labels(labels, state.answer_lang_id)
|
142 |
fig = plotly_express_horizontal_bar_plot(values, translated_labels)
|
143 |
+
st.plotly_chart(fig, use_container_width=True)
|
144 |
|
145 |
|
146 |
st.write(read_markdown("abstract.md"))
|
147 |
st.write(read_markdown("caveats.md"))
|
148 |
st.write("# Methodology")
|
149 |
+
st.image(
|
150 |
+
"./misc/Multilingual-VQA.png", caption="Masked LM model for Image-text Pretraining."
|
151 |
+
)
|
152 |
st.markdown(read_markdown("pretraining.md"))
|
153 |
st.markdown(read_markdown("finetuning.md"))
|
154 |
st.write(read_markdown("challenges.md"))
|
requirements.txt
CHANGED
@@ -4,4 +4,5 @@ git+https://github.com/huggingface/transformers.git
|
|
4 |
torchvision==0.10.0
|
5 |
mtranslate==1.8
|
6 |
black==21.7b0
|
7 |
-
flax==0.3.4
|
|
|
|
4 |
torchvision==0.10.0
|
5 |
mtranslate==1.8
|
6 |
black==21.7b0
|
7 |
+
flax==0.3.4
|
8 |
+
opencv-python==4.5.3
|
translate_answer_mapping.py
CHANGED
@@ -4,6 +4,7 @@ from tqdm import tqdm
|
|
4 |
import ray
|
5 |
from asyncio import Event
|
6 |
from ray.actor import ActorHandle
|
|
|
7 |
ray.init()
|
8 |
from typing import Tuple
|
9 |
|
@@ -48,6 +49,7 @@ class ProgressBarActor:
|
|
48 |
"""
|
49 |
return self.counter
|
50 |
|
|
|
51 |
class ProgressBar:
|
52 |
progress_actor: ActorHandle
|
53 |
total: int
|
@@ -89,14 +91,16 @@ class ProgressBar:
|
|
89 |
with open("answer_reverse_mapping.json") as f:
|
90 |
answer_reverse_mapping = json.load(f)
|
91 |
|
|
|
92 |
@ray.remote
|
93 |
def translate_answer(value, pba):
|
94 |
temp = {}
|
95 |
for lang in ["fr", "es", "de"]:
|
96 |
-
temp.update({lang: translate(value, lang,
|
97 |
pba.update.remote(1)
|
98 |
return temp
|
99 |
|
|
|
100 |
translation_dicts = []
|
101 |
pb = ProgressBar(len(answer_reverse_mapping.values()))
|
102 |
actor = pb.actor
|
@@ -104,8 +108,10 @@ for value in answer_reverse_mapping.values():
|
|
104 |
translation_dicts.append(translate_answer.remote(value, actor))
|
105 |
|
106 |
pb.print_until_done()
|
107 |
-
translation_dict = dict(
|
|
|
|
|
108 |
|
109 |
|
110 |
with open("translation_dict.json", "w") as f:
|
111 |
-
json.dump(translation_dict, f)
|
|
|
4 |
import ray
|
5 |
from asyncio import Event
|
6 |
from ray.actor import ActorHandle
|
7 |
+
|
8 |
ray.init()
|
9 |
from typing import Tuple
|
10 |
|
|
|
49 |
"""
|
50 |
return self.counter
|
51 |
|
52 |
+
|
53 |
class ProgressBar:
|
54 |
progress_actor: ActorHandle
|
55 |
total: int
|
|
|
91 |
with open("answer_reverse_mapping.json") as f:
|
92 |
answer_reverse_mapping = json.load(f)
|
93 |
|
94 |
+
|
95 |
@ray.remote
|
96 |
def translate_answer(value, pba):
|
97 |
temp = {}
|
98 |
for lang in ["fr", "es", "de"]:
|
99 |
+
temp.update({lang: translate(value, lang, "en")})
|
100 |
pba.update.remote(1)
|
101 |
return temp
|
102 |
|
103 |
+
|
104 |
translation_dicts = []
|
105 |
pb = ProgressBar(len(answer_reverse_mapping.values()))
|
106 |
actor = pb.actor
|
|
|
108 |
translation_dicts.append(translate_answer.remote(value, actor))
|
109 |
|
110 |
pb.print_until_done()
|
111 |
+
translation_dict = dict(
|
112 |
+
zip(answer_reverse_mapping.values(), ray.get(translation_dicts))
|
113 |
+
)
|
114 |
|
115 |
|
116 |
with open("translation_dict.json", "w") as f:
|
117 |
+
json.dump(translation_dict, f)
|
utils.py
CHANGED
@@ -7,6 +7,8 @@ from transformers import BertTokenizerFast
|
|
7 |
import plotly.express as px
|
8 |
import json
|
9 |
from PIL import Image
|
|
|
|
|
10 |
class Transform(torch.nn.Module):
|
11 |
def __init__(self, image_size):
|
12 |
super().__init__()
|
@@ -31,7 +33,7 @@ transform = Transform(224)
|
|
31 |
|
32 |
def get_transformed_image(image):
|
33 |
if image.shape[-1] == 3 and isinstance(image, np.ndarray):
|
34 |
-
image = image.transpose(2,0,1)
|
35 |
image = torch.tensor(image)
|
36 |
return transform(image).unsqueeze(0).permute(0, 2, 3, 1).numpy()
|
37 |
|
@@ -49,13 +51,15 @@ def get_top_5_predictions(logits, answer_reverse_mapping):
|
|
49 |
labels = [answer_reverse_mapping[str(i)] for i in indices]
|
50 |
return labels, values
|
51 |
|
52 |
-
|
|
|
53 |
translate_dict = json.load(f)
|
54 |
|
|
|
55 |
def translate_labels(labels, lang_id):
|
56 |
translated_labels = []
|
57 |
for label in labels:
|
58 |
-
if label=="<unk>":
|
59 |
translated_labels.append("<unk>")
|
60 |
elif lang_id == "en":
|
61 |
translated_labels.append(label)
|
@@ -65,5 +69,12 @@ def translate_labels(labels, lang_id):
|
|
65 |
|
66 |
|
67 |
def plotly_express_horizontal_bar_plot(values, labels):
|
68 |
-
fig = px.bar(
|
69 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
import plotly.express as px
|
8 |
import json
|
9 |
from PIL import Image
|
10 |
+
|
11 |
+
|
12 |
class Transform(torch.nn.Module):
|
13 |
def __init__(self, image_size):
|
14 |
super().__init__()
|
|
|
33 |
|
34 |
def get_transformed_image(image):
|
35 |
if image.shape[-1] == 3 and isinstance(image, np.ndarray):
|
36 |
+
image = image.transpose(2, 0, 1)
|
37 |
image = torch.tensor(image)
|
38 |
return transform(image).unsqueeze(0).permute(0, 2, 3, 1).numpy()
|
39 |
|
|
|
51 |
labels = [answer_reverse_mapping[str(i)] for i in indices]
|
52 |
return labels, values
|
53 |
|
54 |
+
|
55 |
+
with open("translation_dict.json") as f:
|
56 |
translate_dict = json.load(f)
|
57 |
|
58 |
+
|
59 |
def translate_labels(labels, lang_id):
|
60 |
translated_labels = []
|
61 |
for label in labels:
|
62 |
+
if label == "<unk>":
|
63 |
translated_labels.append("<unk>")
|
64 |
elif lang_id == "en":
|
65 |
translated_labels.append(label)
|
|
|
69 |
|
70 |
|
71 |
def plotly_express_horizontal_bar_plot(values, labels):
|
72 |
+
fig = px.bar(
|
73 |
+
x=values,
|
74 |
+
y=labels,
|
75 |
+
text=[format(value, ".3%") for value in values],
|
76 |
+
title="Top-5 Predictions",
|
77 |
+
labels={"x": "Scores", "y": "Answers"},
|
78 |
+
orientation="h",
|
79 |
+
)
|
80 |
+
return fig
|