Update app.py
Browse files
app.py
CHANGED
@@ -5,6 +5,10 @@ from rxnim import RXNIM
|
|
5 |
from getReaction import generate_combined_image
|
6 |
import torch
|
7 |
from rxn.reaction import Reaction
|
|
|
|
|
|
|
|
|
8 |
|
9 |
PROMPT_DIR = "prompts/"
|
10 |
ckpt_path = "./rxn/model/model.ckpt"
|
@@ -15,6 +19,7 @@ PROMPT_NAMES = {
|
|
15 |
"2_RxnOCR.txt": "Reaction Image Parsing Workflow",
|
16 |
}
|
17 |
example_diagram = "examples/exp.png"
|
|
|
18 |
|
19 |
def list_prompt_files_with_names():
|
20 |
"""
|
@@ -36,6 +41,7 @@ def parse_reactions(output_json):
|
|
36 |
reactions_data = json.loads(output_json) # 转换 JSON 字符串为字典
|
37 |
reactions_list = reactions_data.get("reactions", [])
|
38 |
detailed_output = []
|
|
|
39 |
|
40 |
for reaction in reactions_list:
|
41 |
reaction_id = reaction.get("reaction_id", "Unknown ID")
|
@@ -50,6 +56,7 @@ def parse_reactions(output_json):
|
|
50 |
]
|
51 |
products = [f"<span style='color:orange'>{p.get('smiles', 'Unknown')}</span>" for p in reaction.get("products", [])]
|
52 |
products_1 = [f"<span style='color:black'>{p.get('smiles', 'Unknown')}</span>" for p in reaction.get("products", [])]
|
|
|
53 |
|
54 |
# 构造反应的完整字符串,定制字体颜色
|
55 |
full_reaction = f"{'.'.join(reactants)}>>{'.'.join(products_1)} | {', '.join(conditions_1)}"
|
@@ -64,7 +71,12 @@ def parse_reactions(output_json):
|
|
64 |
reaction_output += "<br>"
|
65 |
detailed_output.append(reaction_output)
|
66 |
|
67 |
-
|
|
|
|
|
|
|
|
|
|
|
68 |
|
69 |
def process_chem_image(image, selected_task):
|
70 |
chem_mllm = RXNIM()
|
@@ -78,11 +90,12 @@ def process_chem_image(image, selected_task):
|
|
78 |
rxnim_result = chem_mllm.process(image_path, prompt_path)
|
79 |
|
80 |
# 将 JSON 结果解析为结构化输出
|
81 |
-
detailed_reactions = parse_reactions(rxnim_result)
|
82 |
|
83 |
# 调用 RxnScribe 模型处理并生成整合图像
|
84 |
predictions = model.predict_image_file(image_path, molscribe=True, ocr=True)
|
85 |
combined_image_path = generate_combined_image(predictions, image_path)
|
|
|
86 |
|
87 |
json_file_path = "output.json"
|
88 |
with open(json_file_path, "w") as json_file:
|
@@ -90,7 +103,7 @@ def process_chem_image(image, selected_task):
|
|
90 |
|
91 |
|
92 |
# 返回详细反应和整合图像
|
93 |
-
return "\n\n".join(detailed_reactions), combined_image_path, example_diagram, json_file_path
|
94 |
|
95 |
|
96 |
# 获取 prompts 和友好名字
|
@@ -106,26 +119,111 @@ examples = [
|
|
106 |
]
|
107 |
|
108 |
# 定义 Gradio 界面
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
gr.
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
)
|
130 |
-
|
131 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
from getReaction import generate_combined_image
|
6 |
import torch
|
7 |
from rxn.reaction import Reaction
|
8 |
+
from rdkit import Chem
|
9 |
+
from rdkit.Chem import rdChemReactions
|
10 |
+
from rdkit.Chem import Draw
|
11 |
+
|
12 |
|
13 |
PROMPT_DIR = "prompts/"
|
14 |
ckpt_path = "./rxn/model/model.ckpt"
|
|
|
19 |
"2_RxnOCR.txt": "Reaction Image Parsing Workflow",
|
20 |
}
|
21 |
example_diagram = "examples/exp.png"
|
22 |
+
rdkit_image = "examples/image.webp"
|
23 |
|
24 |
def list_prompt_files_with_names():
|
25 |
"""
|
|
|
41 |
reactions_data = json.loads(output_json) # 转换 JSON 字符串为字典
|
42 |
reactions_list = reactions_data.get("reactions", [])
|
43 |
detailed_output = []
|
44 |
+
smiles_output = []
|
45 |
|
46 |
for reaction in reactions_list:
|
47 |
reaction_id = reaction.get("reaction_id", "Unknown ID")
|
|
|
56 |
]
|
57 |
products = [f"<span style='color:orange'>{p.get('smiles', 'Unknown')}</span>" for p in reaction.get("products", [])]
|
58 |
products_1 = [f"<span style='color:black'>{p.get('smiles', 'Unknown')}</span>" for p in reaction.get("products", [])]
|
59 |
+
products_2 = [r.get("smiles", "Unknown") for r in reaction.get("products", [])]
|
60 |
|
61 |
# 构造反应的完整字符串,定制字体颜色
|
62 |
full_reaction = f"{'.'.join(reactants)}>>{'.'.join(products_1)} | {', '.join(conditions_1)}"
|
|
|
71 |
reaction_output += "<br>"
|
72 |
detailed_output.append(reaction_output)
|
73 |
|
74 |
+
reaction_smiles = f"{'.'.join(reactants)}>>{'.'.join(products_2)}"
|
75 |
+
smiles_output.append(reaction_smiles)
|
76 |
+
|
77 |
+
|
78 |
+
|
79 |
+
return detailed_output, smiles_output
|
80 |
|
81 |
def process_chem_image(image, selected_task):
|
82 |
chem_mllm = RXNIM()
|
|
|
90 |
rxnim_result = chem_mllm.process(image_path, prompt_path)
|
91 |
|
92 |
# 将 JSON 结果解析为结构化输出
|
93 |
+
detailed_reactions, smiles_output = parse_reactions(rxnim_result)
|
94 |
|
95 |
# 调用 RxnScribe 模型处理并生成整合图像
|
96 |
predictions = model.predict_image_file(image_path, molscribe=True, ocr=True)
|
97 |
combined_image_path = generate_combined_image(predictions, image_path)
|
98 |
+
#combined_image_path = model.draw_predictions(predictions, image_path)
|
99 |
|
100 |
json_file_path = "output.json"
|
101 |
with open(json_file_path, "w") as json_file:
|
|
|
103 |
|
104 |
|
105 |
# 返回详细反应和整合图像
|
106 |
+
return "\n\n".join(detailed_reactions), smiles_output, combined_image_path, example_diagram, json_file_path
|
107 |
|
108 |
|
109 |
# 获取 prompts 和友好名字
|
|
|
119 |
]
|
120 |
|
121 |
# 定义 Gradio 界面
|
122 |
+
with gr.Blocks() as demo:
|
123 |
+
gr.Markdown(
|
124 |
+
"""
|
125 |
+
|
126 |
+
<center> <h1>Towards Large-scale Chemical Reaction Image Parsing via a Multimodal Large Language Model<h1></center>
|
127 |
+
|
128 |
+
Upload a reaction image and select a predefined task prompt.
|
129 |
+
""")
|
130 |
+
|
131 |
+
|
132 |
+
|
133 |
+
# 上半部分,输入区域
|
134 |
+
with gr.Row(equal_height=False):
|
135 |
+
with gr.Column(scale=1): # 左侧列
|
136 |
+
image_input = gr.Image(type="pil", label="Upload Reaction Image")
|
137 |
+
task_radio = gr.Radio(
|
138 |
+
choices=list(prompts_with_names.keys()),
|
139 |
+
label="Select a predefined task",
|
140 |
+
)
|
141 |
+
with gr.Row(): # Clear 和 Submit 按钮放在同一行
|
142 |
+
clear_button = gr.Button("Clear")
|
143 |
+
process_button = gr.Button("Run", elem_id="submit-btn")
|
144 |
+
|
145 |
+
gr.Markdown("### Reaction Imge Parsing Output")
|
146 |
+
reaction_output = gr.HTML(label="Reaction outputs")
|
147 |
+
|
148 |
+
|
149 |
+
with gr.Column(scale=1):
|
150 |
+
|
151 |
+
gr.Markdown("### Reaction Extraction Output")
|
152 |
+
visualization_output = gr.Image(label="Visualization Output")
|
153 |
+
schematic_diagram = gr.Image(value=example_diagram, label="Schematic Diagram")
|
154 |
+
|
155 |
+
|
156 |
+
with gr.Column(scale=1):
|
157 |
+
gr.Markdown("### Machine-readable Data Output")
|
158 |
+
smiles_output = gr.Textbox(
|
159 |
+
label="Reaction SMILES",
|
160 |
+
show_copy_button=True,
|
161 |
+
interactive=False,
|
162 |
+
visible=False,
|
163 |
+
)
|
164 |
+
|
165 |
+
|
166 |
+
# 下半部分,图像和 JSON 输出
|
167 |
+
@gr.render(inputs = smiles_output) # 使用gr.render修饰器绑定输入和渲染逻辑
|
168 |
+
def show_split(inputs): # 定义处理和展示分割文本的函数
|
169 |
+
if not inputs or isinstance(inputs, str) and inputs.strip() == "": # 检查输入文本是否为空
|
170 |
+
return gr.Textbox(label= f"Reaction SMILES"), gr.Image(value=rdkit_image, label= "RDKit Image generated from Reaction SMILES")
|
171 |
+
else:
|
172 |
+
# 假设输入是逗号分隔的 SMILES 字符串
|
173 |
+
smiles_list = inputs.split(",")
|
174 |
+
smiles_list = [item.strip("[]' ") for item in smiles_list]
|
175 |
+
components = [] # 初始化一个组件列表,用于存放每个 SMILES 对应的 Textbox 组件
|
176 |
+
for i, smiles in enumerate(smiles_list):
|
177 |
+
smiles.replace('"', '').replace("'", "").replace("[", "").replace("]", "")
|
178 |
+
reaction = rdChemReactions.ReactionFromSmarts(smiles)
|
179 |
+
if reaction:
|
180 |
+
img = Draw.ReactionToImage(reaction)
|
181 |
+
components.append(gr.Textbox(value=smiles,label= f"Reaction {i + 1} SMILES", show_copy_button=True, interactive=False))
|
182 |
+
components.append(gr.Image(value=img,label= f"Reaction {i + 1} RDKit Image"))
|
183 |
+
return components # 返回包含所有 SMILES Textbox 组件的列表
|
184 |
+
|
185 |
+
download_json = gr.File(label="Download JSON File",)
|
186 |
+
|
187 |
+
|
188 |
+
|
189 |
+
|
190 |
+
# 示例部分
|
191 |
+
gr.Examples(
|
192 |
+
examples=examples,
|
193 |
+
inputs=[image_input, task_radio],
|
194 |
+
outputs=[reaction_output, smiles_output, visualization_output],
|
195 |
+
)
|
196 |
+
|
197 |
+
# 绑定功能
|
198 |
+
clear_button.click(
|
199 |
+
lambda: (None, None, None, None, None),
|
200 |
+
inputs=[],
|
201 |
+
outputs=[
|
202 |
+
image_input,
|
203 |
+
task_radio,
|
204 |
+
reaction_output,
|
205 |
+
smiles_output,
|
206 |
+
visualization_output,
|
207 |
+
],
|
208 |
+
)
|
209 |
+
|
210 |
+
process_button.click(
|
211 |
+
process_chem_image,
|
212 |
+
inputs=[image_input, task_radio],
|
213 |
+
outputs=[
|
214 |
+
reaction_output,
|
215 |
+
smiles_output,
|
216 |
+
visualization_output,
|
217 |
+
schematic_diagram,
|
218 |
+
download_json,
|
219 |
+
],
|
220 |
+
)
|
221 |
+
|
222 |
+
demo.css = """
|
223 |
+
#submit-btn {
|
224 |
+
background-color: #FF914D;
|
225 |
+
color: white;
|
226 |
+
font-weight: bold;
|
227 |
+
}
|
228 |
+
"""
|
229 |
+
demo.launch()
|