first commit

This commit is contained in:
yadonglu
2024-09-20 15:49:30 -07:00
commit 6cd06a7a86
17 changed files with 2091 additions and 0 deletions

32
README.md Normal file
View File

@@ -0,0 +1,32 @@
# OmniParser: Screen Parsing tool for Pure Vision Based GUI Agent
![Logo](imgs/logo.png)
[![arXiv](https://img.shields.io/badge/Paper-green)](https://arxiv.org/abs/2408.00203)
[![License](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
**OmniParser** is a comprehensive method for parsing user interface screenshots into structured and easy-to-understand elements, which significantly enhances the ability of GPT-4V to generate actions that can be accurately grounded in the corresponding regions of the interface.
## Examples:
We put together a few simple examples in the demo.ipynb.
## Gradio Demo
To run gradio demo, simply run:
```python
python gradion_demo.py
```
## 📚 Citation
Our technical report can be found [here](https://arxiv.org/abs/2408.00203).
If you find our work useful, please consider citing our work:
```
@misc{lu2024omniparserpurevisionbased,
title={OmniParser for Pure Vision Based GUI Agent},
author={Yadong Lu and Jianwei Yang and Yelong Shen and Ahmed Awadallah},
year={2024},
eprint={2408.00203},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2408.00203},
}
```

542
demo.ipynb Normal file

File diff suppressed because one or more lines are too long

104
gradio_demo.py Normal file
View File

@@ -0,0 +1,104 @@
from typing import Optional
import gradio as gr
import numpy as np
import torch
from PIL import Image
import io
import base64, os
from utils import check_ocr_box, get_yolo_model, get_caption_model_processor, get_som_labeled_img
import torch
from PIL import Image
yolo_model = get_yolo_model()
caption_model_processor = get_caption_model_processor('florence', device='cuda') # 'blip2-opt-2.7b-ui', phi3v_ui florence
platform = 'pc'
if platform == 'pc':
draw_bbox_config = {
'text_scale': 0.8,
'text_thickness': 2,
'text_padding': 2,
'thickness': 2,
}
BOX_TRESHOLD = 0.05
elif platform == 'web':
draw_bbox_config = {
'text_scale': 0.8,
'text_thickness': 2,
'text_padding': 3,
'thickness': 3,
}
BOX_TRESHOLD = 0.05
elif platform == 'mobile':
draw_bbox_config = {
'text_scale': 0.8,
'text_thickness': 2,
'text_padding': 3,
'thickness': 3,
}
BOX_TRESHOLD = 0.05
MARKDOWN = """
# OmniParser for Pure Vision Based General GUI Agent 🔥
<div>
<a href="https://arxiv.org/pdf/2408.00203">
<img src="https://img.shields.io/badge/arXiv-2408.00203-b31b1b.svg" alt="Arxiv" style="display:inline-block;">
</a>
</div>
OmniParser is a screen parsing tool to convert general GUI screen to structured elements. **Trained models will be released soon**
"""
DEVICE = torch.device('cuda')
# @spaces.GPU
# @torch.inference_mode()
# @torch.autocast(device_type="cuda", dtype=torch.bfloat16)
def process(
image_input,
prompt: str = None
) -> Optional[Image.Image]:
image_path = "/home/yadonglu/sandbox/data/omniparser_demo/image_input.png"
image_input.save(image_path)
# import pdb; pdb.set_trace()
ocr_bbox_rslt, is_goal_filtered = check_ocr_box(image_path, display_img = False, output_bb_format='xyxy', goal_filtering=None, easyocr_args={'paragraph': False, 'text_threshold':0.9})
text, ocr_bbox = ocr_bbox_rslt
print('prompt:', prompt)
dino_labled_img, label_coordinates, parsed_content_list = get_som_labeled_img(image_path, yolo_model, BOX_TRESHOLD = BOX_TRESHOLD, output_coord_in_ratio=True, ocr_bbox=ocr_bbox,draw_bbox_config=draw_bbox_config, caption_model_processor=caption_model_processor, ocr_text=text,iou_threshold=0.3,prompt=prompt)
image = Image.open(io.BytesIO(base64.b64decode(dino_labled_img)))
print('finish processing')
parsed_content_list = '\n'.join(parsed_content_list)
return image, str(parsed_content_list)
with gr.Blocks() as demo:
gr.Markdown(MARKDOWN)
with gr.Row():
with gr.Column():
image_input_component = gr.Image(
type='pil', label='Upload image')
prompt_input_component = gr.Textbox(label='Prompt', placeholder='')
submit_button_component = gr.Button(
value='Submit', variant='primary')
with gr.Column():
image_output_component = gr.Image(type='pil', label='Image Output')
text_output_component = gr.Textbox(label='Parsed screen elements', placeholder='Text Output')
submit_button_component.click(
fn=process,
inputs=[
image_input_component,
prompt_input_component,
],
outputs=[image_output_component, text_output_component]
)
# demo.launch(debug=False, show_error=True, share=True)
demo.launch(share=True, server_port=7861, server_name='0.0.0.0')

BIN
imgs/logo.png Normal file

Binary file not shown.

After

Width:  |  Height:  |  Size: 3.0 MiB

BIN
imgs/mobile_4.png Normal file

Binary file not shown.

After

Width:  |  Height:  |  Size: 1.8 MiB

BIN
imgs/pc_1.png Normal file

Binary file not shown.

After

Width:  |  Height:  |  Size: 197 KiB

BIN
imgs/settings.png Normal file

Binary file not shown.

After

Width:  |  Height:  |  Size: 805 B

60
omniparser.py Normal file
View File

@@ -0,0 +1,60 @@
from utils import get_som_labeled_img, check_ocr_box, get_caption_model_processor, get_dino_model, get_yolo_model
import torch
from ultralytics import YOLO
from PIL import Image
from typing import Dict, Tuple, List
import io
import base64
config = {
'som_model_path': 'finetuned_icon_detect.pt',
'device': 'cpu',
'caption_model_path': 'Salesforce/blip2-opt-2.7b',
'draw_bbox_config': {
'text_scale': 0.8,
'text_thickness': 2,
'text_padding': 3,
'thickness': 3,
},
'BOX_TRESHOLD': 0.05
}
class Omniparser(object):
def __init__(self, config: Dict):
self.config = config
self.som_model = get_yolo_model(model_path=config['som_model_path'])
# self.caption_model_processor = get_caption_model_processor(config['caption_model_path'], device=cofig['device'])
# self.caption_model_processor['model'].to(torch.float32)
def parse(self, image_path: str):
print('Parsing image:', image_path)
ocr_bbox_rslt, is_goal_filtered = check_ocr_box(image_path, display_img = False, output_bb_format='xyxy', goal_filtering=None, easyocr_args={'paragraph': False, 'text_threshold':0.9})
text, ocr_bbox = ocr_bbox_rslt
draw_bbox_config = self.config['draw_bbox_config']
BOX_TRESHOLD = self.config['BOX_TRESHOLD']
dino_labled_img, label_coordinates, parsed_content_list = get_som_labeled_img(image_path, self.som_model, BOX_TRESHOLD = BOX_TRESHOLD, output_coord_in_ratio=False, ocr_bbox=ocr_bbox,draw_bbox_config=draw_bbox_config, caption_model_processor=None, ocr_text=text,use_local_semantics=False)
image = Image.open(io.BytesIO(base64.b64decode(dino_labled_img)))
# formating output
return_list = [{'from': 'omniparser', 'shape': {'x':coord[0], 'y':coord[1], 'width':coord[2], 'height':coord[3]},
'text': parsed_content_list[i].split(': ')[1], 'type':'text'} for i, (k, coord) in enumerate(label_coordinates.items()) if i < len(parsed_content_list)]
return_list.extend(
[{'from': 'omniparser', 'shape': {'x':coord[0], 'y':coord[1], 'width':coord[2], 'height':coord[3]},
'text': 'None', 'type':'icon'} for i, (k, coord) in enumerate(label_coordinates.items()) if i >= len(parsed_content_list)]
)
return [image, return_list]
parser = Omniparser(config)
image_path = 'examples/pc_1.png'
# time the parser
import time
s = time.time()
image, parsed_content_list = parser.parse(image_path)
device = config['device']
print(f'Time taken for Omniparser on {device}:', time.time() - s)

14
requirement.txt Normal file
View File

@@ -0,0 +1,14 @@
torch==2.2.2
easyocr==1.7.1
torchvision==0.17.2
supervision==0.18.0
openai==1.3.5
transformers==4.40.2
ultralytics==8.1.24
azure-identity
numpy
opencv-python==4.8.1.78
opencv-python-headless==4.8.0.74
supervision==0.18.0
gradio==4.40.0

0
util/__init__.py Executable file
View File

Binary file not shown.

Binary file not shown.

Binary file not shown.

425
util/action_matching.py Normal file
View File

@@ -0,0 +1,425 @@
'''
Adapted from https://github.com/google-research/google-research/tree/master/android_in_the_wild
'''
import jax
import jax.numpy as jnp
import numpy as np
# import action_type as action_type_lib
import enum
class ActionType(enum.IntEnum):
# Placeholders for unused enum values
UNUSED_0 = 0
UNUSED_1 = 1
UNUSED_2 = 2
UNUSED_8 = 8
UNUSED_9 = 9
########### Agent actions ###########
# A type action that sends text to the emulator. Note that this simply sends
# text and does not perform any clicks for element focus or enter presses for
# submitting text.
TYPE = 3
# The dual point action used to represent all gestures.
DUAL_POINT = 4
# These actions differentiate pressing the home and back button from touches.
# They represent explicit presses of back and home performed using ADB.
PRESS_BACK = 5
PRESS_HOME = 6
# An action representing that ADB command for hitting enter was performed.
PRESS_ENTER = 7
########### Episode status actions ###########
# An action used to indicate the desired task has been completed and resets
# the environment. This action should also be used in the case that the task
# has already been completed and there is nothing to do.
# e.g. The task is to turn on the Wi-Fi when it is already on
STATUS_TASK_COMPLETE = 10
# An action used to indicate that desired task is impossible to complete and
# resets the environment. This can be a result of many different things
# including UI changes, Android version differences, etc.
STATUS_TASK_IMPOSSIBLE = 11
_TAP_DISTANCE_THRESHOLD = 0.14 # Fraction of the screen
ANNOTATION_WIDTH_AUGMENT_FRACTION = 1.4
ANNOTATION_HEIGHT_AUGMENT_FRACTION = 1.4
# Interval determining if an action is a tap or a swipe.
_SWIPE_DISTANCE_THRESHOLD = 0.04
def _yx_in_bounding_boxes(
yx, bounding_boxes
):
"""Check if the (y,x) point is contained in each bounding box.
Args:
yx: The (y, x) coordinate in pixels of the point.
bounding_boxes: A 2D int array of shape (num_bboxes, 4), where each row
represents a bounding box: (y_top_left, x_top_left, box_height,
box_width). Note: containment is inclusive of the bounding box edges.
Returns:
is_inside: A 1D bool array where each element specifies if the point is
contained within the respective box.
"""
y, x = yx
# `bounding_boxes` has shape (n_elements, 4); we extract each array along the
# last axis into shape (n_elements, 1), then squeeze unneeded dimension.
top, left, height, width = [
jnp.squeeze(v, axis=-1) for v in jnp.split(bounding_boxes, 4, axis=-1)
]
# The y-axis is inverted for AndroidEnv, so bottom = top + height.
bottom, right = top + height, left + width
return jnp.logical_and(y >= top, y <= bottom) & jnp.logical_and(
x >= left, x <= right)
def _resize_annotation_bounding_boxes(
annotation_positions, annotation_width_augment_fraction,
annotation_height_augment_fraction):
"""Resize the bounding boxes by the given fractions.
Args:
annotation_positions: Array of shape (N, 4), where each row represents the
(y, x, height, width) of the bounding boxes.
annotation_width_augment_fraction: The fraction to augment the box widths,
E.g., 1.4 == 240% total increase.
annotation_height_augment_fraction: Same as described for width, but for box
height.
Returns:
Resized bounding box.
"""
height_change = (
annotation_height_augment_fraction * annotation_positions[:, 2])
width_change = (
annotation_width_augment_fraction * annotation_positions[:, 3])
# Limit bounding box positions to the screen.
resized_annotations = jnp.stack([
jnp.maximum(0, annotation_positions[:, 0] - (height_change / 2)),
jnp.maximum(0, annotation_positions[:, 1] - (width_change / 2)),
jnp.minimum(1, annotation_positions[:, 2] + height_change),
jnp.minimum(1, annotation_positions[:, 3] + width_change),
],
axis=1)
return resized_annotations
def is_tap_action(normalized_start_yx,
normalized_end_yx):
distance = jnp.linalg.norm(
jnp.array(normalized_start_yx) - jnp.array(normalized_end_yx))
return distance <= _SWIPE_DISTANCE_THRESHOLD
def _is_non_dual_point_action(action_type):
return jnp.not_equal(action_type, ActionType.DUAL_POINT)
def _check_tap_actions_match(
tap_1_yx,
tap_2_yx,
annotation_positions,
matching_tap_distance_threshold_screen_percentage,
annotation_width_augment_fraction,
annotation_height_augment_fraction,
):
"""Determines if two tap actions are the same."""
resized_annotation_positions = _resize_annotation_bounding_boxes(
annotation_positions,
annotation_width_augment_fraction,
annotation_height_augment_fraction,
)
# Check if the ground truth tap action falls in an annotation's bounding box.
tap1_in_box = _yx_in_bounding_boxes(tap_1_yx, resized_annotation_positions)
tap2_in_box = _yx_in_bounding_boxes(tap_2_yx, resized_annotation_positions)
both_in_box = jnp.max(tap1_in_box & tap2_in_box)
# If the ground-truth tap action falls outside any of the annotation
# bounding boxes or one of the actions is inside a bounding box and the other
# is outside bounding box or vice versa, compare the points using Euclidean
# distance.
within_threshold = (
jnp.linalg.norm(jnp.array(tap_1_yx) - jnp.array(tap_2_yx))
<= matching_tap_distance_threshold_screen_percentage
)
return jnp.logical_or(both_in_box, within_threshold)
def _check_drag_actions_match(
drag_1_touch_yx,
drag_1_lift_yx,
drag_2_touch_yx,
drag_2_lift_yx,
):
"""Determines if two drag actions are the same."""
# Store drag deltas (the change in the y and x coordinates from touch to
# lift), magnitudes, and the index of the main axis, which is the axis with
# the greatest change in coordinate value (e.g. a drag starting at (0, 0) and
# ending at (0.3, 0.5) has a main axis index of 1).
drag_1_deltas = drag_1_lift_yx - drag_1_touch_yx
drag_1_magnitudes = jnp.abs(drag_1_deltas)
drag_1_main_axis = np.argmax(drag_1_magnitudes)
drag_2_deltas = drag_2_lift_yx - drag_2_touch_yx
drag_2_magnitudes = jnp.abs(drag_2_deltas)
drag_2_main_axis = np.argmax(drag_2_magnitudes)
return jnp.equal(drag_1_main_axis, drag_2_main_axis)
def check_actions_match(
action_1_touch_yx,
action_1_lift_yx,
action_1_action_type,
action_2_touch_yx,
action_2_lift_yx,
action_2_action_type,
annotation_positions,
tap_distance_threshold = _TAP_DISTANCE_THRESHOLD,
annotation_width_augment_fraction = ANNOTATION_WIDTH_AUGMENT_FRACTION,
annotation_height_augment_fraction = ANNOTATION_HEIGHT_AUGMENT_FRACTION,
):
"""Determines if two actions are considered to be the same.
Two actions being "the same" is defined here as two actions that would result
in a similar screen state.
Args:
action_1_touch_yx: The (y, x) coordinates of the first action's touch.
action_1_lift_yx: The (y, x) coordinates of the first action's lift.
action_1_action_type: The action type of the first action.
action_2_touch_yx: The (y, x) coordinates of the second action's touch.
action_2_lift_yx: The (y, x) coordinates of the second action's lift.
action_2_action_type: The action type of the second action.
annotation_positions: The positions of the UI annotations for the screen. It
is A 2D int array of shape (num_bboxes, 4), where each row represents a
bounding box: (y_top_left, x_top_left, box_height, box_width). Note that
containment is inclusive of the bounding box edges.
tap_distance_threshold: The threshold that determines if two taps result in
a matching screen state if they don't fall the same bounding boxes.
annotation_width_augment_fraction: The fraction to increase the width of the
bounding box by.
annotation_height_augment_fraction: The fraction to increase the height of
of the bounding box by.
Returns:
A boolean representing whether the two given actions are the same or not.
"""
action_1_touch_yx = jnp.asarray(action_1_touch_yx)
action_1_lift_yx = jnp.asarray(action_1_lift_yx)
action_2_touch_yx = jnp.asarray(action_2_touch_yx)
action_2_lift_yx = jnp.asarray(action_2_lift_yx)
# Checks if at least one of the actions is global (i.e. not DUAL_POINT),
# because if that is the case, only the actions' types need to be compared.
has_non_dual_point_action = jnp.logical_or(
_is_non_dual_point_action(action_1_action_type),
_is_non_dual_point_action(action_2_action_type),
)
#print("non dual point: "+str(has_non_dual_point_action))
different_dual_point_types = jnp.logical_xor(
is_tap_action(action_1_touch_yx, action_1_lift_yx),
is_tap_action(action_2_touch_yx, action_2_lift_yx),
)
#print("different dual type: "+str(different_dual_point_types))
is_tap = jnp.logical_and(
is_tap_action(action_1_touch_yx, action_1_lift_yx),
is_tap_action(action_2_touch_yx, action_2_lift_yx),
)
#print("is tap: "+str(is_tap))
taps_match = _check_tap_actions_match(
action_1_touch_yx,
action_2_touch_yx,
annotation_positions,
tap_distance_threshold,
annotation_width_augment_fraction,
annotation_height_augment_fraction,
)
#print("tap match: "+str(taps_match))
taps_match = jnp.logical_and(is_tap, taps_match)
#print("tap match: "+str(taps_match))
drags_match = _check_drag_actions_match(
action_1_touch_yx, action_1_lift_yx, action_2_touch_yx, action_2_lift_yx
)
drags_match = jnp.where(is_tap, False, drags_match)
#print("drag match: "+str(drags_match))
return jnp.where(
has_non_dual_point_action,
jnp.equal(action_1_action_type, action_2_action_type),
jnp.where(
different_dual_point_types,
False,
jnp.logical_or(taps_match, drags_match),
),
)
def action_2_format(step_data):
# 把test数据集中的动作格式转换为计算matching score的格式
action_type = step_data["action_type_id"]
if action_type == 4:
if step_data["action_type_text"] == 'click': # 点击
touch_point = step_data["touch"]
lift_point = step_data["lift"]
else: # 上下左右滑动
if step_data["action_type_text"] == 'scroll down':
touch_point = [0.5, 0.8]
lift_point = [0.5, 0.2]
elif step_data["action_type_text"] == 'scroll up':
touch_point = [0.5, 0.2]
lift_point = [0.5, 0.8]
elif step_data["action_type_text"] == 'scroll left':
touch_point = [0.2, 0.5]
lift_point = [0.8, 0.5]
elif step_data["action_type_text"] == 'scroll right':
touch_point = [0.8, 0.5]
lift_point = [0.2, 0.5]
else:
touch_point = [-1.0, -1.0]
lift_point = [-1.0, -1.0]
if action_type == 3:
typed_text = step_data["type_text"]
else:
typed_text = ""
action = {"action_type": action_type, "touch_point": touch_point, "lift_point": lift_point,
"typed_text": typed_text}
action["touch_point"] = [action["touch_point"][1], action["touch_point"][0]]
action["lift_point"] = [action["lift_point"][1], action["lift_point"][0]]
action["typed_text"] = action["typed_text"].lower()
return action
def pred_2_format(step_data):
# 把模型输出的内容转换为计算action_matching的格式
action_type = step_data["action_type"]
if action_type == 4: # 点击
action_type_new = 4
touch_point = step_data["click_point"]
lift_point = step_data["click_point"]
typed_text = ""
elif action_type == 0:
action_type_new = 4
touch_point = [0.5, 0.8]
lift_point = [0.5, 0.2]
typed_text = ""
elif action_type == 1:
action_type_new = 4
touch_point = [0.5, 0.2]
lift_point = [0.5, 0.8]
typed_text = ""
elif action_type == 8:
action_type_new = 4
touch_point = [0.2, 0.5]
lift_point = [0.8, 0.5]
typed_text = ""
elif action_type == 9:
action_type_new = 4
touch_point = [0.8, 0.5]
lift_point = [0.2, 0.5]
typed_text = ""
else:
action_type_new = action_type
touch_point = [-1.0, -1.0]
lift_point = [-1.0, -1.0]
typed_text = ""
if action_type_new == 3:
typed_text = step_data["typed_text"]
action = {"action_type": action_type_new, "touch_point": touch_point, "lift_point": lift_point,
"typed_text": typed_text}
action["touch_point"] = [action["touch_point"][1], action["touch_point"][0]]
action["lift_point"] = [action["lift_point"][1], action["lift_point"][0]]
action["typed_text"] = action["typed_text"].lower()
return action
def pred_2_format_simplified(step_data):
# 把模型输出的内容转换为计算action_matching的格式
action_type = step_data["action_type"]
if action_type == 'click' : # 点击
action_type_new = 4
touch_point = step_data["click_point"]
lift_point = step_data["click_point"]
typed_text = ""
elif action_type == 'scroll' and step_data["direction"] == 'down':
action_type_new = 4
touch_point = [0.5, 0.8]
lift_point = [0.5, 0.2]
typed_text = ""
elif action_type == 'scroll' and step_data["direction"] == 'up':
action_type_new = 4
touch_point = [0.5, 0.2]
lift_point = [0.5, 0.8]
typed_text = ""
elif action_type == 'scroll' and step_data["direction"] == 'left':
action_type_new = 4
touch_point = [0.2, 0.5]
lift_point = [0.8, 0.5]
typed_text = ""
elif action_type == 'scroll' and step_data["direction"] == 'right':
action_type_new = 4
touch_point = [0.8, 0.5]
lift_point = [0.2, 0.5]
typed_text = ""
elif action_type == 'type':
action_type_new = 3
touch_point = [-1.0, -1.0]
lift_point = [-1.0, -1.0]
typed_text = step_data["text"]
elif action_type == 'navigate_back':
action_type_new = 5
touch_point = [-1.0, -1.0]
lift_point = [-1.0, -1.0]
typed_text = ""
elif action_type == 'navigate_home':
action_type_new = 6
touch_point = [-1.0, -1.0]
lift_point = [-1.0, -1.0]
typed_text = ""
else:
action_type_new = action_type
touch_point = [-1.0, -1.0]
lift_point = [-1.0, -1.0]
typed_text = ""
# if action_type_new == 'type':
# typed_text = step_data["text"]
action = {"action_type": action_type_new, "touch_point": touch_point, "lift_point": lift_point,
"typed_text": typed_text}
action["touch_point"] = [action["touch_point"][1], action["touch_point"][0]]
action["lift_point"] = [action["lift_point"][1], action["lift_point"][0]]
action["typed_text"] = action["typed_text"].lower()
return action

45
util/action_type.py Normal file
View File

@@ -0,0 +1,45 @@
'''
Adapted from https://github.com/google-research/google-research/tree/master/android_in_the_wild
'''
import enum
class ActionType(enum.IntEnum):
# Placeholders for unused enum values
UNUSED_0 = 0
UNUSED_1 = 1
UNUSED_2 = 2
UNUSED_8 = 8
UNUSED_9 = 9
########### Agent actions ###########
# A type action that sends text to the emulator. Note that this simply sends
# text and does not perform any clicks for element focus or enter presses for
# submitting text.
TYPE = 3
# The dual point action used to represent all gestures.
DUAL_POINT = 4
# These actions differentiate pressing the home and back button from touches.
# They represent explicit presses of back and home performed using ADB.
PRESS_BACK = 5
PRESS_HOME = 6
# An action representing that ADB command for hitting enter was performed.
PRESS_ENTER = 7
########### Episode status actions ###########
# An action used to indicate the desired task has been completed and resets
# the environment. This action should also be used in the case that the task
# has already been completed and there is nothing to do.
# e.g. The task is to turn on the Wi-Fi when it is already on
STATUS_TASK_COMPLETE = 10
# An action used to indicate that desired task is impossible to complete and
# resets the environment. This can be a result of many different things
# including UI changes, Android version differences, etc.
STATUS_TASK_IMPOSSIBLE = 11

262
util/box_annotator.py Normal file
View File

@@ -0,0 +1,262 @@
from typing import List, Optional, Union, Tuple
import cv2
import numpy as np
from supervision.detection.core import Detections
from supervision.draw.color import Color, ColorPalette
class BoxAnnotator:
"""
A class for drawing bounding boxes on an image using detections provided.
Attributes:
color (Union[Color, ColorPalette]): The color to draw the bounding box,
can be a single color or a color palette
thickness (int): The thickness of the bounding box lines, default is 2
text_color (Color): The color of the text on the bounding box, default is white
text_scale (float): The scale of the text on the bounding box, default is 0.5
text_thickness (int): The thickness of the text on the bounding box,
default is 1
text_padding (int): The padding around the text on the bounding box,
default is 5
"""
def __init__(
self,
color: Union[Color, ColorPalette] = ColorPalette.DEFAULT,
thickness: int = 3, # 1 for seeclick 2 for mind2web and 3 for demo
text_color: Color = Color.BLACK,
text_scale: float = 0.5, # 0.8 for mobile/web, 0.3 for desktop # 0.4 for mind2web
text_thickness: int = 2, #1, # 2 for demo
text_padding: int = 10,
avoid_overlap: bool = True,
):
self.color: Union[Color, ColorPalette] = color
self.thickness: int = thickness
self.text_color: Color = text_color
self.text_scale: float = text_scale
self.text_thickness: int = text_thickness
self.text_padding: int = text_padding
self.avoid_overlap: bool = avoid_overlap
def annotate(
self,
scene: np.ndarray,
detections: Detections,
labels: Optional[List[str]] = None,
skip_label: bool = False,
image_size: Optional[Tuple[int, int]] = None,
) -> np.ndarray:
"""
Draws bounding boxes on the frame using the detections provided.
Args:
scene (np.ndarray): The image on which the bounding boxes will be drawn
detections (Detections): The detections for which the
bounding boxes will be drawn
labels (Optional[List[str]]): An optional list of labels
corresponding to each detection. If `labels` are not provided,
corresponding `class_id` will be used as label.
skip_label (bool): Is set to `True`, skips bounding box label annotation.
Returns:
np.ndarray: The image with the bounding boxes drawn on it
Example:
```python
import supervision as sv
classes = ['person', ...]
image = ...
detections = sv.Detections(...)
box_annotator = sv.BoxAnnotator()
labels = [
f"{classes[class_id]} {confidence:0.2f}"
for _, _, confidence, class_id, _ in detections
]
annotated_frame = box_annotator.annotate(
scene=image.copy(),
detections=detections,
labels=labels
)
```
"""
font = cv2.FONT_HERSHEY_SIMPLEX
for i in range(len(detections)):
x1, y1, x2, y2 = detections.xyxy[i].astype(int)
class_id = (
detections.class_id[i] if detections.class_id is not None else None
)
idx = class_id if class_id is not None else i
color = (
self.color.by_idx(idx)
if isinstance(self.color, ColorPalette)
else self.color
)
cv2.rectangle(
img=scene,
pt1=(x1, y1),
pt2=(x2, y2),
color=color.as_bgr(),
thickness=self.thickness,
)
if skip_label:
continue
text = (
f"{class_id}"
if (labels is None or len(detections) != len(labels))
else labels[i]
)
text_width, text_height = cv2.getTextSize(
text=text,
fontFace=font,
fontScale=self.text_scale,
thickness=self.text_thickness,
)[0]
if not self.avoid_overlap:
text_x = x1 + self.text_padding
text_y = y1 - self.text_padding
text_background_x1 = x1
text_background_y1 = y1 - 2 * self.text_padding - text_height
text_background_x2 = x1 + 2 * self.text_padding + text_width
text_background_y2 = y1
# text_x = x1 - self.text_padding - text_width
# text_y = y1 + self.text_padding + text_height
# text_background_x1 = x1 - 2 * self.text_padding - text_width
# text_background_y1 = y1
# text_background_x2 = x1
# text_background_y2 = y1 + 2 * self.text_padding + text_height
else:
text_x, text_y, text_background_x1, text_background_y1, text_background_x2, text_background_y2 = get_optimal_label_pos(self.text_padding, text_width, text_height, x1, y1, x2, y2, detections, image_size)
cv2.rectangle(
img=scene,
pt1=(text_background_x1, text_background_y1),
pt2=(text_background_x2, text_background_y2),
color=color.as_bgr(),
thickness=cv2.FILLED,
)
# import pdb; pdb.set_trace()
box_color = color.as_rgb()
luminance = 0.299 * box_color[0] + 0.587 * box_color[1] + 0.114 * box_color[2]
text_color = (0,0,0) if luminance > 160 else (255,255,255)
cv2.putText(
img=scene,
text=text,
org=(text_x, text_y),
fontFace=font,
fontScale=self.text_scale,
# color=self.text_color.as_rgb(),
color=text_color,
thickness=self.text_thickness,
lineType=cv2.LINE_AA,
)
return scene
def box_area(box):
return (box[2] - box[0]) * (box[3] - box[1])
def intersection_area(box1, box2):
x1 = max(box1[0], box2[0])
y1 = max(box1[1], box2[1])
x2 = min(box1[2], box2[2])
y2 = min(box1[3], box2[3])
return max(0, x2 - x1) * max(0, y2 - y1)
def IoU(box1, box2, return_max=True):
intersection = intersection_area(box1, box2)
union = box_area(box1) + box_area(box2) - intersection
if box_area(box1) > 0 and box_area(box2) > 0:
ratio1 = intersection / box_area(box1)
ratio2 = intersection / box_area(box2)
else:
ratio1, ratio2 = 0, 0
if return_max:
return max(intersection / union, ratio1, ratio2)
else:
return intersection / union
def get_optimal_label_pos(text_padding, text_width, text_height, x1, y1, x2, y2, detections, image_size):
""" check overlap of text and background detection box, and get_optimal_label_pos,
pos: str, position of the text, must be one of 'top left', 'top right', 'outer left', 'outer right' TODO: if all are overlapping, return the last one, i.e. outer right
Threshold: default to 0.3
"""
def get_is_overlap(detections, text_background_x1, text_background_y1, text_background_x2, text_background_y2, image_size):
is_overlap = False
for i in range(len(detections)):
detection = detections.xyxy[i].astype(int)
if IoU([text_background_x1, text_background_y1, text_background_x2, text_background_y2], detection) > 0.3:
is_overlap = True
break
# check if the text is out of the image
if text_background_x1 < 0 or text_background_x2 > image_size[0] or text_background_y1 < 0 or text_background_y2 > image_size[1]:
is_overlap = True
return is_overlap
# if pos == 'top left':
text_x = x1 + text_padding
text_y = y1 - text_padding
text_background_x1 = x1
text_background_y1 = y1 - 2 * text_padding - text_height
text_background_x2 = x1 + 2 * text_padding + text_width
text_background_y2 = y1
is_overlap = get_is_overlap(detections, text_background_x1, text_background_y1, text_background_x2, text_background_y2, image_size)
if not is_overlap:
return text_x, text_y, text_background_x1, text_background_y1, text_background_x2, text_background_y2
# elif pos == 'outer left':
text_x = x1 - text_padding - text_width
text_y = y1 + text_padding + text_height
text_background_x1 = x1 - 2 * text_padding - text_width
text_background_y1 = y1
text_background_x2 = x1
text_background_y2 = y1 + 2 * text_padding + text_height
is_overlap = get_is_overlap(detections, text_background_x1, text_background_y1, text_background_x2, text_background_y2, image_size)
if not is_overlap:
return text_x, text_y, text_background_x1, text_background_y1, text_background_x2, text_background_y2
# elif pos == 'outer right':
text_x = x2 + text_padding
text_y = y1 + text_padding + text_height
text_background_x1 = x2
text_background_y1 = y1
text_background_x2 = x2 + 2 * text_padding + text_width
text_background_y2 = y1 + 2 * text_padding + text_height
is_overlap = get_is_overlap(detections, text_background_x1, text_background_y1, text_background_x2, text_background_y2, image_size)
if not is_overlap:
return text_x, text_y, text_background_x1, text_background_y1, text_background_x2, text_background_y2
# elif pos == 'top right':
text_x = x2 - text_padding - text_width
text_y = y1 - text_padding
text_background_x1 = x2 - 2 * text_padding - text_width
text_background_y1 = y1 - 2 * text_padding - text_height
text_background_x2 = x2
text_background_y2 = y1
is_overlap = get_is_overlap(detections, text_background_x1, text_background_y1, text_background_x2, text_background_y2, image_size)
if not is_overlap:
return text_x, text_y, text_background_x1, text_background_y1, text_background_x2, text_background_y2
return text_x, text_y, text_background_x1, text_background_y1, text_background_x2, text_background_y2

607
utils.py Executable file
View File

@@ -0,0 +1,607 @@
# from ultralytics import YOLO
import os
import io
import base64
import time
from PIL import Image, ImageDraw, ImageFont
import json
import requests
# utility function
import os
from openai import AzureOpenAI
import json
import sys
import os
import cv2
import numpy as np
# %matplotlib inline
from matplotlib import pyplot as plt
import easyocr
reader = easyocr.Reader(['en']) # this needs to run only once to load the model into memory # 'ch_sim',
import time
import base64
import os
import ast
import torch
from typing import Tuple, List
from torchvision.ops import box_convert
import re
from torchvision.transforms import ToPILImage
import supervision as sv
import torchvision.transforms as T
def get_caption_model_processor(model_name="Salesforce/blip2-opt-2.7b", device=None):
if not device:
device = "cuda" if torch.cuda.is_available() else "cpu"
if model_name == "Salesforce/blip2-opt-2.7b":
from transformers import Blip2Processor, Blip2ForConditionalGeneration
processor = Blip2Processor.from_pretrained("Salesforce/blip2-opt-2.7b")
model = Blip2ForConditionalGeneration.from_pretrained(
"Salesforce/blip2-opt-2.7b", device_map=None, torch_dtype=torch.float16
# '/home/yadonglu/sandbox/data/orca/blipv2_ui_merge', device_map=None, torch_dtype=torch.float16
)
elif model_name == "blip2-opt-2.7b-ui":
from transformers import Blip2Processor, Blip2ForConditionalGeneration
processor = Blip2Processor.from_pretrained("Salesforce/blip2-opt-2.7b")
if device == 'cpu':
model = Blip2ForConditionalGeneration.from_pretrained(
'/home/yadonglu/sandbox/data/orca/blipv2_ui_merge', device_map=None, torch_dtype=torch.float32
)
else:
model = Blip2ForConditionalGeneration.from_pretrained(
'/home/yadonglu/sandbox/data/orca/blipv2_ui_merge', device_map=None, torch_dtype=torch.float16
)
elif model_name == "florence":
from transformers import AutoProcessor, AutoModelForCausalLM
processor = AutoProcessor.from_pretrained("microsoft/Florence-2-base", trust_remote_code=True)
if device == 'cpu':
model = AutoModelForCausalLM.from_pretrained("/home/yadonglu/sandbox/data/orca/florence-2-base-ft-fft_ep1_rai", torch_dtype=torch.float32, trust_remote_code=True)#.to(device)
else:
model = AutoModelForCausalLM.from_pretrained("/home/yadonglu/sandbox/data/orca/florence-2-base-ft-fft_ep1_rai_win_ep5_fixed", torch_dtype=torch.float16, trust_remote_code=True).to(device)
elif model_name == 'phi3v_ui':
from transformers import AutoModelForCausalLM, AutoProcessor
model_id = "microsoft/Phi-3-vision-128k-instruct"
model = AutoModelForCausalLM.from_pretrained('/home/yadonglu/sandbox/data/orca/phi3v_ui', device_map=device, trust_remote_code=True, torch_dtype="auto")
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
elif model_name == 'phi3v':
from transformers import AutoModelForCausalLM, AutoProcessor
model_id = "microsoft/Phi-3-vision-128k-instruct"
model = AutoModelForCausalLM.from_pretrained(model_id, device_map=device, trust_remote_code=True, torch_dtype="auto")
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
return {'model': model.to(device), 'processor': processor}
def get_yolo_model():
from ultralytics import YOLO
# Load the model.
# model = YOLO('/home/yadonglu/sandbox/data/yolo/runs/detect/yolov8n_v8_xcyc/weights/best.pt')
model = YOLO('/home/yadonglu/sandbox/data/yolo/runs/detect/yolov8n_v8_seq_xcyc_b32_n4_office_ep20/weights/best.pt')
return model
def get_parsed_content_icon(filtered_boxes, ocr_bbox, image_source, caption_model_processor, prompt=None):
to_pil = ToPILImage()
if ocr_bbox:
non_ocr_boxes = filtered_boxes[len(ocr_bbox):]
else:
non_ocr_boxes = filtered_boxes
croped_pil_image = []
for i, coord in enumerate(non_ocr_boxes):
xmin, xmax = int(coord[0]*image_source.shape[1]), int(coord[2]*image_source.shape[1])
ymin, ymax = int(coord[1]*image_source.shape[0]), int(coord[3]*image_source.shape[0])
cropped_image = image_source[ymin:ymax, xmin:xmax, :]
croped_pil_image.append(to_pil(cropped_image))
# import pdb; pdb.set_trace()
model, processor = caption_model_processor['model'], caption_model_processor['processor']
if not prompt:
if 'florence' in model.config.name_or_path:
prompt = "<CAPTION>"
else:
prompt = "The image shows"
# prompt = "NO gender!NO gender!NO gender! The image shows a icon:"
batch_size = 10 # Number of samples per batch
generated_texts = []
device = model.device
for i in range(0, len(croped_pil_image), batch_size):
batch = croped_pil_image[i:i+batch_size]
if model.device.type == 'cuda':
inputs = processor(images=batch, text=[prompt]*len(batch), return_tensors="pt").to(device=device, dtype=torch.float16)
else:
inputs = processor(images=batch, text=[prompt]*len(batch), return_tensors="pt").to(device=device)
if 'florence' in model.config.name_or_path:
generated_ids = model.generate(input_ids=inputs["input_ids"],pixel_values=inputs["pixel_values"],max_new_tokens=1024,num_beams=3, do_sample=False)
else:
generated_ids = model.generate(**inputs, max_length=100, num_beams=5, no_repeat_ngram_size=2, early_stopping=True, num_return_sequences=1) # temperature=0.01, do_sample=True,
generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)
generated_text = [gen.strip() for gen in generated_text]
generated_texts.extend(generated_text)
return generated_texts
def get_parsed_content_icon_phi3v(filtered_boxes, ocr_bbox, image_source, caption_model_processor):
to_pil = ToPILImage()
if ocr_bbox:
non_ocr_boxes = filtered_boxes[len(ocr_bbox):]
else:
non_ocr_boxes = filtered_boxes
croped_pil_image = []
for i, coord in enumerate(non_ocr_boxes):
xmin, xmax = int(coord[0]*image_source.shape[1]), int(coord[2]*image_source.shape[1])
ymin, ymax = int(coord[1]*image_source.shape[0]), int(coord[3]*image_source.shape[0])
cropped_image = image_source[ymin:ymax, xmin:xmax, :]
croped_pil_image.append(to_pil(cropped_image))
model, processor = caption_model_processor['model'], caption_model_processor['processor']
device = model.device
messages = [{"role": "user", "content": "<|image_1|>\ndescribe the icon in one sentence"}]
prompt = processor.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
batch_size = 5 # Number of samples per batch
generated_texts = []
for i in range(0, len(croped_pil_image), batch_size):
images = croped_pil_image[i:i+batch_size]
image_inputs = [processor.image_processor(x, return_tensors="pt") for x in images]
inputs ={'input_ids': [], 'attention_mask': [], 'pixel_values': [], 'image_sizes': []}
texts = [prompt] * len(images)
for i, txt in enumerate(texts):
input = processor._convert_images_texts_to_inputs(image_inputs[i], txt, return_tensors="pt")
inputs['input_ids'].append(input['input_ids'])
inputs['attention_mask'].append(input['attention_mask'])
inputs['pixel_values'].append(input['pixel_values'])
inputs['image_sizes'].append(input['image_sizes'])
max_len = max([x.shape[1] for x in inputs['input_ids']])
for i, v in enumerate(inputs['input_ids']):
inputs['input_ids'][i] = torch.cat([processor.tokenizer.pad_token_id * torch.ones(1, max_len - v.shape[1], dtype=torch.long), v], dim=1)
inputs['attention_mask'][i] = torch.cat([torch.zeros(1, max_len - v.shape[1], dtype=torch.long), inputs['attention_mask'][i]], dim=1)
inputs_cat = {k: torch.concatenate(v).to(device) for k, v in inputs.items()}
generation_args = {
"max_new_tokens": 25,
"temperature": 0.01,
"do_sample": False,
}
generate_ids = model.generate(**inputs_cat, eos_token_id=processor.tokenizer.eos_token_id, **generation_args)
# # remove input tokens
generate_ids = generate_ids[:, inputs_cat['input_ids'].shape[1]:]
response = processor.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)
response = [res.strip('\n').strip() for res in response]
generated_texts.extend(response)
return generated_texts
def remove_overlap(boxes, iou_threshold, ocr_bbox=None):
assert ocr_bbox is None or isinstance(ocr_bbox, List)
def box_area(box):
return (box[2] - box[0]) * (box[3] - box[1])
def intersection_area(box1, box2):
x1 = max(box1[0], box2[0])
y1 = max(box1[1], box2[1])
x2 = min(box1[2], box2[2])
y2 = min(box1[3], box2[3])
return max(0, x2 - x1) * max(0, y2 - y1)
def IoU(box1, box2):
intersection = intersection_area(box1, box2)
union = box_area(box1) + box_area(box2) - intersection + 1e-6
if box_area(box1) > 0 and box_area(box2) > 0:
ratio1 = intersection / box_area(box1)
ratio2 = intersection / box_area(box2)
else:
ratio1, ratio2 = 0, 0
return max(intersection / union, ratio1, ratio2)
boxes = boxes.tolist()
filtered_boxes = []
if ocr_bbox:
filtered_boxes.extend(ocr_bbox)
# print('ocr_bbox!!!', ocr_bbox)
for i, box1 in enumerate(boxes):
# if not any(IoU(box1, box2) > iou_threshold and box_area(box1) > box_area(box2) for j, box2 in enumerate(boxes) if i != j):
is_valid_box = True
for j, box2 in enumerate(boxes):
if i != j and IoU(box1, box2) > iou_threshold and box_area(box1) > box_area(box2):
is_valid_box = False
break
if is_valid_box:
# add the following 2 lines to include ocr bbox
if ocr_bbox:
if not any(IoU(box1, box3) > iou_threshold for k, box3 in enumerate(ocr_bbox)):
filtered_boxes.append(box1)
else:
filtered_boxes.append(box1)
return torch.tensor(filtered_boxes)
def load_image(image_path: str) -> Tuple[np.array, torch.Tensor]:
transform = T.Compose(
[
T.RandomResize([800], max_size=1333),
T.ToTensor(),
T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
]
)
image_source = Image.open(image_path).convert("RGB")
image = np.asarray(image_source)
image_transformed, _ = transform(image_source, None)
return image, image_transformed
def annotate(image_source: np.ndarray, boxes: torch.Tensor, logits: torch.Tensor, phrases: List[str], text_scale: float,
text_padding=5, text_thickness=2, thickness=3) -> np.ndarray:
"""
This function annotates an image with bounding boxes and labels.
Parameters:
image_source (np.ndarray): The source image to be annotated.
boxes (torch.Tensor): A tensor containing bounding box coordinates. in cxcywh format, pixel scale
logits (torch.Tensor): A tensor containing confidence scores for each bounding box.
phrases (List[str]): A list of labels for each bounding box.
text_scale (float): The scale of the text to be displayed. 0.8 for mobile/web, 0.3 for desktop # 0.4 for mind2web
Returns:
np.ndarray: The annotated image.
"""
h, w, _ = image_source.shape
boxes = boxes * torch.Tensor([w, h, w, h])
xyxy = box_convert(boxes=boxes, in_fmt="cxcywh", out_fmt="xyxy").numpy()
xywh = box_convert(boxes=boxes, in_fmt="cxcywh", out_fmt="xywh").numpy()
detections = sv.Detections(xyxy=xyxy)
labels = [f"{phrase}" for phrase in range(boxes.shape[0])]
from util.box_annotator import BoxAnnotator
box_annotator = BoxAnnotator(text_scale=text_scale, text_padding=text_padding,text_thickness=text_thickness,thickness=thickness) # 0.8 for mobile/web, 0.3 for desktop # 0.4 for mind2web
annotated_frame = image_source.copy()
annotated_frame = box_annotator.annotate(scene=annotated_frame, detections=detections, labels=labels, image_size=(w,h))
label_coordinates = {f"{phrase}": v for phrase, v in zip(phrases, xywh)}
return annotated_frame, label_coordinates
def predict(model, image, caption, box_threshold, text_threshold):
""" Use huggingface model to replace the original model
"""
model, processor = model['model'], model['processor']
device = model.device
inputs = processor(images=image, text=caption, return_tensors="pt").to(device)
with torch.no_grad():
outputs = model(**inputs)
results = processor.post_process_grounded_object_detection(
outputs,
inputs.input_ids,
box_threshold=box_threshold, # 0.4,
text_threshold=text_threshold, # 0.3,
target_sizes=[image.size[::-1]]
)[0]
boxes, logits, phrases = results["boxes"], results["scores"], results["labels"]
return boxes, logits, phrases
def predict_yolo(model, image_path, box_threshold):
""" Use huggingface model to replace the original model
"""
# model = model['model']
result = model.predict(
source=image_path,
conf=box_threshold,
# iou=0.5, # default 0.7
)
boxes = result[0].boxes.xyxy#.tolist() # in pixel space
conf = result[0].boxes.conf
phrases = [str(i) for i in range(len(boxes))]
return boxes, conf, phrases
def get_som_labeled_img(img_path, model=None, BOX_TRESHOLD = 0.01, output_coord_in_ratio=False, ocr_bbox=None, text_scale=0.4, text_padding=5, draw_bbox_config=None, caption_model_processor=None, ocr_text=[], use_local_semantics=True, iou_threshold=0.9,prompt=None):
""" ocr_bbox: list of xyxy format bbox
"""
TEXT_PROMPT = "clickable buttons on the screen"
# BOX_TRESHOLD = 0.02 # 0.05/0.02 for web and 0.1 for mobile
TEXT_TRESHOLD = 0.01 # 0.9 # 0.01
image_source = Image.open(img_path).convert("RGB")
w, h = image_source.size
# import pdb; pdb.set_trace()
if False: # TODO
xyxy, logits, phrases = predict(model=model, image=image_source, caption=TEXT_PROMPT, box_threshold=BOX_TRESHOLD, text_threshold=TEXT_TRESHOLD)
else:
xyxy, logits, phrases = predict_yolo(model=model, image_path=img_path, box_threshold=BOX_TRESHOLD)
xyxy = xyxy / torch.Tensor([w, h, w, h]).to(xyxy.device)
image_source = np.asarray(image_source)
phrases = [str(i) for i in range(len(phrases))]
# annotate the image with labels
h, w, _ = image_source.shape
if ocr_bbox:
ocr_bbox = torch.tensor(ocr_bbox) / torch.Tensor([w, h, w, h])
ocr_bbox=ocr_bbox.tolist()
else:
print('no ocr bbox!!!')
ocr_bbox = None
filtered_boxes = remove_overlap(boxes=xyxy, iou_threshold=iou_threshold, ocr_bbox=ocr_bbox)
# get parsed icon local semantics
if use_local_semantics:
caption_model = caption_model_processor['model']
if 'phi3_v' in caption_model.config.model_type:
parsed_content_icon = get_parsed_content_icon_phi3v(filtered_boxes, ocr_bbox, image_source, caption_model_processor)
else:
parsed_content_icon = get_parsed_content_icon(filtered_boxes, ocr_bbox, image_source, caption_model_processor, prompt=prompt)
ocr_text = [f"Text Box ID {i}: {txt}" for i, txt in enumerate(ocr_text)]
icon_start = len(ocr_text)
parsed_content_icon_ls = []
for i, txt in enumerate(parsed_content_icon):
parsed_content_icon_ls.append(f"Icon Box ID {str(i+icon_start)}: {txt}")
parsed_content_merged = ocr_text + parsed_content_icon_ls
else:
ocr_text = [f"Text Box ID {i}: {txt}" for i, txt in enumerate(ocr_text)]
parsed_content_merged = ocr_text
filtered_boxes = box_convert(boxes=filtered_boxes, in_fmt="xyxy", out_fmt="cxcywh")
phrases = [i for i in range(len(filtered_boxes))]
# draw boxes
if draw_bbox_config:
annotated_frame, label_coordinates = annotate(image_source=image_source, boxes=filtered_boxes, logits=logits, phrases=phrases, **draw_bbox_config)
else:
annotated_frame, label_coordinates = annotate(image_source=image_source, boxes=filtered_boxes, logits=logits, phrases=phrases, text_scale=text_scale, text_padding=text_padding)
pil_img = Image.fromarray(annotated_frame)
buffered = io.BytesIO()
pil_img.save(buffered, format="PNG")
encoded_image = base64.b64encode(buffered.getvalue()).decode('ascii')
if output_coord_in_ratio:
# h, w, _ = image_source.shape
label_coordinates = {k: [v[0]/w, v[1]/h, v[2]/w, v[3]/h] for k, v in label_coordinates.items()}
assert w == annotated_frame.shape[1] and h == annotated_frame.shape[0]
return encoded_image, label_coordinates, parsed_content_merged
def get_xywh(input):
x, y, w, h = input[0][0], input[0][1], input[2][0] - input[0][0], input[2][1] - input[0][1]
x, y, w, h = int(x), int(y), int(w), int(h)
return x, y, w, h
def get_xyxy(input):
x, y, xp, yp = input[0][0], input[0][1], input[2][0], input[2][1]
x, y, xp, yp = int(x), int(y), int(xp), int(yp)
return x, y, xp, yp
def get_xywh_yolo(input):
x, y, w, h = input[0], input[1], input[2] - input[0], input[3] - input[1]
x, y, w, h = int(x), int(y), int(w), int(h)
return x, y, w, h
def run_api(body, max_tokens=1024):
'''
API call, check https://platform.openai.com/docs/guides/vision for the latest api usage.
'''
max_num_trial = 3
num_trial = 0
while num_trial < max_num_trial:
try:
response = client.chat.completions.create(
model=deployment,
messages=body,
temperature=0.01,
max_tokens=max_tokens,
)
return response.choices[0].message.content
except:
print('retry call gptv', num_trial)
num_trial += 1
time.sleep(10)
return ''
def call_gpt4v_new(message_text, image_path=None, max_tokens=2048):
if image_path:
try:
with open(image_path, "rb") as img_file:
encoded_image = base64.b64encode(img_file.read()).decode('ascii')
except:
encoded_image = image_path
if image_path:
content = [{"type": "image_url","image_url": {"url": f"data:image/jpeg;base64,{encoded_image}"}}, {"type": "text","text": message_text},]
else:
content = [{"type": "text","text": message_text},]
max_num_trial = 3
num_trial = 0
call_api_success = True
while num_trial < max_num_trial:
try:
response = client.chat.completions.create(
model=deployment,
messages=[
{
"role": "system",
"content": [
{
"type": "text",
"text": "You are an AI assistant that is good at making plans and analyzing screens, and helping people find information."
},
]
},
{
"role": "user",
"content": content
}
],
temperature=0.01,
max_tokens=max_tokens,
)
ans_1st_pass = response.choices[0].message.content
break
except:
print('retry call gptv', num_trial)
num_trial += 1
ans_1st_pass = ''
time.sleep(10)
if num_trial == max_num_trial:
call_api_success = False
return ans_1st_pass, call_api_success
def check_ocr_box(image_path, display_img = True, output_bb_format='xywh', goal_filtering=None, easyocr_args=None):
if easyocr_args is None:
easyocr_args = {}
result = reader.readtext(image_path, **easyocr_args)
is_goal_filtered = False
if goal_filtering:
ocr_filter_fs = "Example 1:\n Based on task and ocr results, ```In summary, the task related bboxes are: [([[3060, 111], [3135, 111], [3135, 141], [3060, 141]], 'Share', 0.949013667261589), ([[3068, 197], [3135, 197], [3135, 227], [3068, 227]], 'Link _', 0.3567054243152049), ([[3006, 321], [3178, 321], [3178, 354], [3006, 354]], 'Manage Access', 0.8800734456437066)] ``` \n Example 2:\n Based on task and ocr results, ```In summary, the task related bboxes are: [([[3060, 111], [3135, 111], [3135, 141], [3060, 141]], 'Search Google or type a URL', 0.949013667261589)] ```"
# message_text = f"Based on the ocr results which contains text+bounding box in a dictionary, please filter it so that it only contains the task related bboxes. The task is: {goal_filtering}, the ocr results are: {str(result)}. Your final answer should be in the exact same format as the ocr results, please do not include any other redundant information, please do not include any analysis."
message_text = f"Based on the task and ocr results which contains text+bounding box in a dictionary, please filter it so that it only contains the task related bboxes. Requirement: 1. first give a brief analysis. 2. provide an answer in the format: ```In summary, the task related bboxes are: ..```, you must put it inside ``` ```. Do not include any info after ```.\n {ocr_filter_fs}\n The task is: {goal_filtering}, the ocr results are: {str(result)}."
prompt = [{"role":"system", "content": "You are an AI assistant that helps people find the correct way to operate computer or smartphone."}, {"role":"user","content": message_text},]
print('[Perform OCR filtering by goal] ongoing ...')
# pred, _, _ = call_gpt4(prompt)
pred, _, = call_gpt4v(message_text)
# import pdb; pdb.set_trace()
try:
# match = re.search(r"```(.*?)```", pred, re.DOTALL)
# result = match.group(1).strip()
# pred = result.split('In summary, the task related bboxes are:')[-1].strip()
pred = pred.split('In summary, the task related bboxes are:')[-1].strip().strip('```')
result = ast.literal_eval(pred)
print('[Perform OCR filtering by goal] success!!! Filtered buttons: ', pred)
is_goal_filtered = True
except:
print('[Perform OCR filtering by goal] failed or unused!!!')
pass
# added_prompt = [{"role":"assistant","content":pred},
# {"role":"user","content": "given the previous answers, please provide the final answer in the exact same format as the ocr results, please do not include any other redundant information, please do not include any analysis."}]
# prompt.extend(added_prompt)
# pred, _, _ = call_gpt4(prompt)
# print('goal filtering pred 2nd:', pred)
# result = ast.literal_eval(pred)
# print('goal filtering pred:', result[-5:])
coord = [item[0] for item in result]
text = [item[1] for item in result]
# confidence = [item[2] for item in result]
# if confidence_filtering:
# coord = [coord[i] for i in range(len(coord)) if confidence[i] > confidence_filtering]
# text = [text[i] for i in range(len(text)) if confidence[i] > confidence_filtering]
# read the image using cv2
if display_img:
opencv_img = cv2.imread(image_path)
opencv_img = cv2.cvtColor(opencv_img, cv2.COLOR_RGB2BGR)
bb = []
for item in coord:
x, y, a, b = get_xywh(item)
# print(x, y, a, b)
bb.append((x, y, a, b))
cv2.rectangle(opencv_img, (x, y), (x+a, y+b), (0, 255, 0), 2)
# Display the image
plt.imshow(opencv_img)
else:
if output_bb_format == 'xywh':
bb = [get_xywh(item) for item in coord]
elif output_bb_format == 'xyxy':
bb = [get_xyxy(item) for item in coord]
# print('bounding box!!!', bb)
return (text, bb), is_goal_filtered
def get_pred_gptv(message_text, yolo_labled_img, label_coordinates, summarize_history=True, verbose=True, history=None, id_key='Click ID'):
""" This func first
1. call gptv(yolo_labled_img, text bbox+task) -> ans_1st_cal
2. call gpt4(ans_1st_cal, label_coordinates) -> final ans
"""
# Configuration
encoded_image = yolo_labled_img
# Payload for the request
if not history:
messages = [
{"role": "system", "content": [{"type": "text","text": "You are an AI assistant that is great at interpreting screenshot and predict action."},]},
{"role": "user","content": [{"type": "text","text": message_text}, {"type": "image_url","image_url": {"url": f"data:image/jpeg;base64,{encoded_image}"}},]}
]
else:
messages = [
{"role": "system", "content": [{"type": "text","text": "You are an AI assistant that is great at interpreting screenshot and predict action."},]},
history,
{"role": "user","content": [{"type": "image_url","image_url": {"url": f"data:image/jpeg;base64,{encoded_image}"}},{"type": "text","text": message_text},]}
]
payload = {
"messages": messages,
"temperature": 0.01, # 0.01
"top_p": 0.95,
"max_tokens": 800
}
max_num_trial = 3
num_trial = 0
call_api_success = True
while num_trial < max_num_trial:
try:
# response = requests.post(GPT4V_ENDPOINT, headers=headers, json=payload)
# response.raise_for_status() # Will raise an HTTPError if the HTTP request returned an unsuccessful status code
# ans_1st_pass = response.json()['choices'][0]['message']['content']
response = client.chat.completions.create(
model=deployment,
messages=messages,
temperature=0.01,
max_tokens=512,
)
ans_1st_pass = response.choices[0].message.content
break
except requests.RequestException as e:
print('retry call gptv', num_trial)
num_trial += 1
ans_1st_pass = ''
time.sleep(30)
# raise SystemExit(f"Failed to make the request. Error: {e}")
if num_trial == max_num_trial:
call_api_success = False
if verbose:
print('Answer by GPTV: ', ans_1st_pass)
# extract by simple parsing
try:
match = re.search(r"```(.*?)```", ans_1st_pass, re.DOTALL)
if match:
result = match.group(1).strip()
pred = result.split('In summary, the next action I will perform is:')[-1].strip().replace('\\', '')
pred = ast.literal_eval(pred)
else:
pred = ans_1st_pass.split('In summary, the next action I will perform is:')[-1].strip().replace('\\', '')
pred = ast.literal_eval(pred)
if id_key in pred:
icon_id = pred[id_key]
bbox = label_coordinates[str(icon_id)]
pred['click_point'] = [bbox[0] + bbox[2]/2, bbox[1] + bbox[3]/2]
except:
# import pdb; pdb.set_trace()
print('gptv action regex extract fail!!!')
print('ans_1st_pass:', ans_1st_pass)
pred = {'action_type': 'CLICK', 'click_point': [0, 0], 'value': 'None', 'is_completed': False}
step_pred_summary = None
if summarize_history:
step_pred_summary, _ = call_gpt4v_new('Summarize what action you decide to perform in the current step, in one sentence, and do not include any icon box number: ' + ans_1st_pass, max_tokens=128)
print('step_pred_summary', step_pred_summary)
return pred, [call_api_success, ans_1st_pass, None, step_pred_summary]
# return pred, [call_api_success, message_2nd, completion_2nd.choices[0].message.content, step_pred_summary]