Spaces:
Runtime error
Runtime error
removed BROS model & Adding OmniParser
Browse filesmain changes include:
OmniParser Integration:
Added YOLO model loading for icon detection
Added Florence-2 model for captioning
Proper handling of both models in the pipeline
Analysis Pipeline:
Object detection with YOLO
Caption generation for detected elements
Structured output with bounding boxes and descriptions
User Interface:
Updated model information
Added UI-specific strengths and capabilities
Proper debug information for UI parsing
app.py
CHANGED
|
@@ -1,20 +1,20 @@
|
|
| 1 |
import streamlit as st
|
| 2 |
from PIL import Image
|
| 3 |
import torch
|
| 4 |
-
import json
|
| 5 |
from transformers import (
|
| 6 |
DonutProcessor,
|
| 7 |
VisionEncoderDecoderModel,
|
| 8 |
LayoutLMv3Processor,
|
| 9 |
LayoutLMv3ForSequenceClassification,
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
LlavaProcessor,
|
| 13 |
-
LlavaForConditionalGeneration
|
| 14 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
from datetime import datetime
|
| 16 |
|
| 17 |
-
# Cache the model loading to improve performance
|
| 18 |
@st.cache_resource
|
| 19 |
def load_model(model_name):
|
| 20 |
"""Load the selected model and processor"""
|
|
@@ -31,13 +31,21 @@ def load_model(model_name):
|
|
| 31 |
processor = LayoutLMv3Processor.from_pretrained("microsoft/layoutlmv3-base")
|
| 32 |
model = LayoutLMv3ForSequenceClassification.from_pretrained("microsoft/layoutlmv3-base")
|
| 33 |
|
| 34 |
-
elif model_name == "
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 41 |
|
| 42 |
return model, processor
|
| 43 |
except Exception as e:
|
|
@@ -47,14 +55,54 @@ def load_model(model_name):
|
|
| 47 |
def analyze_document(image, model_name, model, processor):
|
| 48 |
"""Analyze document using selected model"""
|
| 49 |
try:
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 53 |
pixel_values = processor(image, return_tensors="pt").pixel_values
|
| 54 |
task_prompt = "<s_cord>analyze the document and extract information</s_cord>"
|
| 55 |
decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
|
| 56 |
|
| 57 |
-
# Generate output with improved parameters
|
| 58 |
outputs = model.generate(
|
| 59 |
pixel_values,
|
| 60 |
decoder_input_ids=decoder_input_ids,
|
|
@@ -68,31 +116,41 @@ def analyze_document(image, model_name, model, processor):
|
|
| 68 |
return_dict_in_generate=True
|
| 69 |
)
|
| 70 |
|
| 71 |
-
# Process and clean the output
|
| 72 |
sequence = processor.batch_decode(outputs.sequences)[0]
|
| 73 |
sequence = sequence.replace(task_prompt, "").replace("</s_cord>", "").strip()
|
| 74 |
|
| 75 |
-
# Try to parse as JSON, fallback to raw text
|
| 76 |
try:
|
| 77 |
result = json.loads(sequence)
|
| 78 |
except json.JSONDecodeError:
|
| 79 |
result = {"raw_text": sequence}
|
| 80 |
|
| 81 |
elif model_name == "LayoutLMv3":
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
|
|
|
|
|
|
|
|
|
|
| 85 |
|
| 86 |
-
|
| 87 |
-
|
| 88 |
-
|
| 89 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 90 |
|
| 91 |
-
elif model_name == "LLaVA-1.5":
|
| 92 |
-
inputs = processor(image, return_tensors="pt")
|
| 93 |
-
outputs = model.generate(**inputs, max_length=256)
|
| 94 |
-
result = {"generated_text": processor.decode(outputs[0], skip_special_tokens=True)}
|
| 95 |
-
|
| 96 |
return result
|
| 97 |
|
| 98 |
except Exception as e:
|
|
@@ -157,26 +215,18 @@ with col2:
|
|
| 157 |
"Donut": {
|
| 158 |
"description": "Best for structured OCR and document format understanding",
|
| 159 |
"memory": "6-8GB",
|
| 160 |
-
"strengths": ["Structured OCR", "Memory efficient", "Good with fixed formats"]
|
| 161 |
-
"best_for": ["Invoices", "Forms", "Structured documents"]
|
| 162 |
},
|
| 163 |
"LayoutLMv3": {
|
| 164 |
"description": "Strong layout understanding with reasoning capabilities",
|
| 165 |
"memory": "12-15GB",
|
| 166 |
-
"strengths": ["Layout understanding", "Reasoning", "Pre-trained knowledge"]
|
| 167 |
-
"best_for": ["Complex layouts", "Mixed content", "Tables"]
|
| 168 |
-
},
|
| 169 |
-
"BROS": {
|
| 170 |
-
"description": "Memory efficient with fast inference",
|
| 171 |
-
"memory": "4-6GB",
|
| 172 |
-
"strengths": ["Fast inference", "Memory efficient", "Easy fine-tuning"],
|
| 173 |
-
"best_for": ["Simple documents", "Quick analysis", "Basic OCR"]
|
| 174 |
},
|
| 175 |
-
"
|
| 176 |
-
"description": "
|
| 177 |
-
"memory": "
|
| 178 |
-
"strengths": ["
|
| 179 |
-
"best_for": ["
|
| 180 |
}
|
| 181 |
}
|
| 182 |
|
|
|
|
| 1 |
import streamlit as st
|
| 2 |
from PIL import Image
|
| 3 |
import torch
|
|
|
|
| 4 |
from transformers import (
|
| 5 |
DonutProcessor,
|
| 6 |
VisionEncoderDecoderModel,
|
| 7 |
LayoutLMv3Processor,
|
| 8 |
LayoutLMv3ForSequenceClassification,
|
| 9 |
+
AutoProcessor,
|
| 10 |
+
AutoModelForCausalLM
|
|
|
|
|
|
|
| 11 |
)
|
| 12 |
+
from ultralytics import YOLO
|
| 13 |
+
import io
|
| 14 |
+
import base64
|
| 15 |
+
import json
|
| 16 |
from datetime import datetime
|
| 17 |
|
|
|
|
| 18 |
@st.cache_resource
|
| 19 |
def load_model(model_name):
|
| 20 |
"""Load the selected model and processor"""
|
|
|
|
| 31 |
processor = LayoutLMv3Processor.from_pretrained("microsoft/layoutlmv3-base")
|
| 32 |
model = LayoutLMv3ForSequenceClassification.from_pretrained("microsoft/layoutlmv3-base")
|
| 33 |
|
| 34 |
+
elif model_name == "OmniParser":
|
| 35 |
+
# Load YOLO model for icon detection
|
| 36 |
+
yolo_model = YOLO('microsoft/OmniParser', task='detect')
|
| 37 |
+
# Load Florence-2 model for captioning
|
| 38 |
+
processor = AutoProcessor.from_pretrained("microsoft/Florence-2-base", trust_remote_code=True)
|
| 39 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 40 |
+
"microsoft/OmniParser",
|
| 41 |
+
torch_dtype=torch.float16,
|
| 42 |
+
trust_remote_code=True
|
| 43 |
+
)
|
| 44 |
+
return {
|
| 45 |
+
'yolo': yolo_model,
|
| 46 |
+
'processor': processor,
|
| 47 |
+
'model': model
|
| 48 |
+
}
|
| 49 |
|
| 50 |
return model, processor
|
| 51 |
except Exception as e:
|
|
|
|
| 55 |
def analyze_document(image, model_name, model, processor):
|
| 56 |
"""Analyze document using selected model"""
|
| 57 |
try:
|
| 58 |
+
if model_name == "OmniParser":
|
| 59 |
+
# Save image temporarily
|
| 60 |
+
temp_path = "temp_image.png"
|
| 61 |
+
image.save(temp_path)
|
| 62 |
+
|
| 63 |
+
# Configure box detection parameters
|
| 64 |
+
box_threshold = 0.05 # Can be made configurable
|
| 65 |
+
iou_threshold = 0.1 # Can be made configurable
|
| 66 |
+
|
| 67 |
+
# Run YOLO detection
|
| 68 |
+
yolo_results = model['yolo'](
|
| 69 |
+
temp_path,
|
| 70 |
+
conf=box_threshold,
|
| 71 |
+
iou=iou_threshold,
|
| 72 |
+
device='cpu' if not torch.cuda.is_available() else 'cuda'
|
| 73 |
+
)
|
| 74 |
+
|
| 75 |
+
# Process detections
|
| 76 |
+
results = []
|
| 77 |
+
for det in yolo_results[0].boxes.data:
|
| 78 |
+
x1, y1, x2, y2, conf, cls = det
|
| 79 |
+
|
| 80 |
+
# Get region of interest
|
| 81 |
+
roi = image.crop((x1, y1, x2, y2))
|
| 82 |
+
|
| 83 |
+
# Generate caption using Florence-2
|
| 84 |
+
inputs = processor(images=roi, return_tensors="pt")
|
| 85 |
+
outputs = model['model'].generate(**inputs, max_length=50)
|
| 86 |
+
caption = processor.decode(outputs[0], skip_special_tokens=True)
|
| 87 |
+
|
| 88 |
+
results.append({
|
| 89 |
+
"bbox": [float(x) for x in [x1, y1, x2, y2]],
|
| 90 |
+
"confidence": float(conf),
|
| 91 |
+
"class": int(cls),
|
| 92 |
+
"caption": caption
|
| 93 |
+
})
|
| 94 |
+
|
| 95 |
+
return {
|
| 96 |
+
"detected_elements": len(results),
|
| 97 |
+
"elements": results
|
| 98 |
+
}
|
| 99 |
+
|
| 100 |
+
# [Previous model handling remains the same...]
|
| 101 |
+
elif model_name == "Donut":
|
| 102 |
pixel_values = processor(image, return_tensors="pt").pixel_values
|
| 103 |
task_prompt = "<s_cord>analyze the document and extract information</s_cord>"
|
| 104 |
decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
|
| 105 |
|
|
|
|
| 106 |
outputs = model.generate(
|
| 107 |
pixel_values,
|
| 108 |
decoder_input_ids=decoder_input_ids,
|
|
|
|
| 116 |
return_dict_in_generate=True
|
| 117 |
)
|
| 118 |
|
|
|
|
| 119 |
sequence = processor.batch_decode(outputs.sequences)[0]
|
| 120 |
sequence = sequence.replace(task_prompt, "").replace("</s_cord>", "").strip()
|
| 121 |
|
|
|
|
| 122 |
try:
|
| 123 |
result = json.loads(sequence)
|
| 124 |
except json.JSONDecodeError:
|
| 125 |
result = {"raw_text": sequence}
|
| 126 |
|
| 127 |
elif model_name == "LayoutLMv3":
|
| 128 |
+
encoded_inputs = processor(
|
| 129 |
+
image,
|
| 130 |
+
return_tensors="pt",
|
| 131 |
+
add_special_tokens=True,
|
| 132 |
+
return_offsets_mapping=True
|
| 133 |
+
)
|
| 134 |
|
| 135 |
+
outputs = model(**encoded_inputs)
|
| 136 |
+
predictions = outputs.logits.argmax(-1).squeeze().tolist()
|
| 137 |
+
|
| 138 |
+
words = processor.tokenizer.convert_ids_to_tokens(
|
| 139 |
+
encoded_inputs.input_ids.squeeze().tolist()
|
| 140 |
+
)
|
| 141 |
+
|
| 142 |
+
result = {
|
| 143 |
+
"predictions": [
|
| 144 |
+
{
|
| 145 |
+
"text": word,
|
| 146 |
+
"label": pred
|
| 147 |
+
}
|
| 148 |
+
for word, pred in zip(words, predictions)
|
| 149 |
+
if word not in ["<s>", "</s>", "<pad>"]
|
| 150 |
+
],
|
| 151 |
+
"confidence_scores": outputs.logits.softmax(-1).max(-1).values.squeeze().tolist()
|
| 152 |
+
}
|
| 153 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 154 |
return result
|
| 155 |
|
| 156 |
except Exception as e:
|
|
|
|
| 215 |
"Donut": {
|
| 216 |
"description": "Best for structured OCR and document format understanding",
|
| 217 |
"memory": "6-8GB",
|
| 218 |
+
"strengths": ["Structured OCR", "Memory efficient", "Good with fixed formats"]
|
|
|
|
| 219 |
},
|
| 220 |
"LayoutLMv3": {
|
| 221 |
"description": "Strong layout understanding with reasoning capabilities",
|
| 222 |
"memory": "12-15GB",
|
| 223 |
+
"strengths": ["Layout understanding", "Reasoning", "Pre-trained knowledge"]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 224 |
},
|
| 225 |
+
"OmniParser": {
|
| 226 |
+
"description": "General screen parsing tool for UI understanding",
|
| 227 |
+
"memory": "8-10GB",
|
| 228 |
+
"strengths": ["UI element detection", "Interactive element recognition", "Function description"],
|
| 229 |
+
"best_for": ["Screenshots", "UI analysis", "Interactive elements"]
|
| 230 |
}
|
| 231 |
}
|
| 232 |
|