Update my_model/detector/object_detection.py
Browse files
my_model/detector/object_detection.py
CHANGED
@@ -1,4 +1,4 @@
|
|
1 |
-
|
2 |
import streamlit as st
|
3 |
from transformers import AutoImageProcessor, AutoModelForObjectDetection
|
4 |
import torch
|
@@ -11,18 +11,18 @@ from my_model.utilities.gen_utilities import get_image_path, get_model_path ,sho
|
|
11 |
|
12 |
class ObjectDetector:
|
13 |
"""
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
|
25 |
-
def __init__(self):
|
26 |
"""
|
27 |
Initializes the ObjectDetector class with default values.
|
28 |
"""
|
@@ -33,17 +33,17 @@ class ObjectDetector:
|
|
33 |
self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
34 |
|
35 |
|
36 |
-
def load_model(self, model_name='detic', pretrained=True, model_version='yolov5s'):
|
37 |
"""
|
38 |
-
|
39 |
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
|
45 |
-
|
46 |
-
|
47 |
"""
|
48 |
|
49 |
self.model_name = model_name
|
@@ -55,12 +55,15 @@ class ObjectDetector:
|
|
55 |
raise ValueError(f"Unsupported model name: {model_name}")
|
56 |
|
57 |
|
58 |
-
def _load_detic_model(self, pretrained):
|
59 |
"""
|
60 |
Load the Detic model.
|
61 |
|
62 |
Args:
|
63 |
pretrained (bool): If True, load a pretrained model.
|
|
|
|
|
|
|
64 |
"""
|
65 |
|
66 |
try:
|
@@ -72,13 +75,15 @@ class ObjectDetector:
|
|
72 |
raise
|
73 |
|
74 |
|
75 |
-
def _load_yolov5_model(self, pretrained, model_version):
|
76 |
"""
|
77 |
Load the YOLOv5 model.
|
78 |
|
79 |
Args:
|
80 |
pretrained (bool): If True, load a pretrained model.
|
81 |
model_version (str): Version of the YOLOv5 model.
|
|
|
|
|
82 |
"""
|
83 |
|
84 |
try:
|
@@ -92,13 +97,16 @@ class ObjectDetector:
|
|
92 |
raise
|
93 |
|
94 |
|
95 |
-
def process_image(self, image_input):
|
96 |
"""
|
97 |
Process the image from the given path or file-like object.
|
|
|
98 |
Args:
|
99 |
-
image_input (str
|
|
|
100 |
Returns:
|
101 |
Image.Image: Processed image in RGB format.
|
|
|
102 |
Raises:
|
103 |
Exception: If an error occurs during image processing.
|
104 |
"""
|
@@ -119,16 +127,17 @@ class ObjectDetector:
|
|
119 |
raise
|
120 |
|
121 |
|
122 |
-
|
123 |
"""
|
124 |
Detect objects in the given image using the loaded model.
|
125 |
|
126 |
Args:
|
127 |
image (Image.Image): Image in which to detect objects.
|
128 |
-
threshold (float): Model detection confidence.
|
129 |
|
130 |
Returns:
|
131 |
-
|
|
|
132 |
|
133 |
Raises:
|
134 |
ValueError: If the model is not loaded or the model name is unsupported.
|
@@ -142,7 +151,7 @@ class ObjectDetector:
|
|
142 |
raise ValueError("Model not loaded or unsupported model name")
|
143 |
|
144 |
|
145 |
-
def _detect_with_detic(self, image, threshold):
|
146 |
"""
|
147 |
Detect objects using the Detic model.
|
148 |
|
@@ -151,8 +160,8 @@ class ObjectDetector:
|
|
151 |
threshold (float): The confidence threshold for detections.
|
152 |
|
153 |
Returns:
|
154 |
-
|
155 |
-
|
156 |
"""
|
157 |
|
158 |
inputs = self.processor(images=image, return_tensors="pt")
|
@@ -171,7 +180,7 @@ class ObjectDetector:
|
|
171 |
return detected_objects_str, detected_objects_list
|
172 |
|
173 |
|
174 |
-
def _detect_with_yolov5(self, image, threshold):
|
175 |
"""
|
176 |
Detect objects using the YOLOv5 model.
|
177 |
|
@@ -180,8 +189,8 @@ class ObjectDetector:
|
|
180 |
threshold (float): The confidence threshold for detections.
|
181 |
|
182 |
Returns:
|
183 |
-
|
184 |
-
|
185 |
"""
|
186 |
|
187 |
cv2_img = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
|
@@ -198,13 +207,13 @@ class ObjectDetector:
|
|
198 |
return detected_objects_str, detected_objects_list
|
199 |
|
200 |
|
201 |
-
def draw_boxes(self, image, detected_objects, show_confidence=True):
|
202 |
"""
|
203 |
Draw bounding boxes around detected objects in the image.
|
204 |
|
205 |
Args:
|
206 |
image (Image.Image): Image on which to draw.
|
207 |
-
detected_objects (
|
208 |
show_confidence (bool): Whether to show confidence scores.
|
209 |
|
210 |
Returns:
|
@@ -232,7 +241,7 @@ class ObjectDetector:
|
|
232 |
return image
|
233 |
|
234 |
|
235 |
-
def detect_and_draw_objects(image_path, model_type='yolov5', threshold=0.2, show_confidence=True):
|
236 |
"""
|
237 |
Detects objects in an image, draws bounding boxes around them, and returns the processed image and a string description.
|
238 |
|
@@ -243,7 +252,7 @@ def detect_and_draw_objects(image_path, model_type='yolov5', threshold=0.2, show
|
|
243 |
show_confidence (bool): Whether to show confidence scores on the output image.
|
244 |
|
245 |
Returns:
|
246 |
-
|
247 |
"""
|
248 |
|
249 |
detector = ObjectDetector()
|
@@ -252,9 +261,3 @@ def detect_and_draw_objects(image_path, model_type='yolov5', threshold=0.2, show
|
|
252 |
detected_objects_string, detected_objects_list = detector.detect_objects(image, threshold=threshold)
|
253 |
image_with_boxes = detector.draw_boxes(image, detected_objects_list, show_confidence=show_confidence)
|
254 |
return image_with_boxes, detected_objects_string
|
255 |
-
|
256 |
-
|
257 |
-
|
258 |
-
if __name__ == "__main__":
|
259 |
-
pass
|
260 |
-
|
|
|
1 |
+
from typing import Union, Optional, List, Tuple
|
2 |
import streamlit as st
|
3 |
from transformers import AutoImageProcessor, AutoModelForObjectDetection
|
4 |
import torch
|
|
|
11 |
|
12 |
class ObjectDetector:
|
13 |
"""
|
14 |
+
A class for detecting objects in images using models like Detic and YOLOv5.
|
15 |
+
This class supports loading and using different object detection models to identify objects
|
16 |
+
in images and draw bounding boxes around them.
|
17 |
+
|
18 |
+
Attributes:
|
19 |
+
model (torch.nn.Module or None): The loaded object detection model.
|
20 |
+
processor (transformers.AutoImageProcessor or None): Processor for the Detic model.
|
21 |
+
model_name (str or None): Name of the model used for detection.
|
22 |
+
device (str): Device to use for computation ('cuda' if available, otherwise 'cpu').
|
23 |
+
"""
|
24 |
|
25 |
+
def __init__(self) -> None:
|
26 |
"""
|
27 |
Initializes the ObjectDetector class with default values.
|
28 |
"""
|
|
|
33 |
self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
34 |
|
35 |
|
36 |
+
def load_model(self, model_name: str = 'detic', pretrained: bool = True, model_version: str = 'yolov5s') -> None:
|
37 |
"""
|
38 |
+
Load the specified object detection model.
|
39 |
|
40 |
+
Args:
|
41 |
+
model_name (str): Name of the model to load. Options are 'detic' and 'yolov5'.
|
42 |
+
pretrained (bool): Boolean indicating if a pretrained model should be used.
|
43 |
+
model_version (str): Version of the YOLOv5 model, applicable only when using YOLOv5.
|
44 |
|
45 |
+
Raises:
|
46 |
+
ValueError: If an unsupported model name is provided.
|
47 |
"""
|
48 |
|
49 |
self.model_name = model_name
|
|
|
55 |
raise ValueError(f"Unsupported model name: {model_name}")
|
56 |
|
57 |
|
58 |
+
def _load_detic_model(self, pretrained: bool) -> None:
|
59 |
"""
|
60 |
Load the Detic model.
|
61 |
|
62 |
Args:
|
63 |
pretrained (bool): If True, load a pretrained model.
|
64 |
+
|
65 |
+
Raises:
|
66 |
+
Exception: If an error occurs during model loading.
|
67 |
"""
|
68 |
|
69 |
try:
|
|
|
75 |
raise
|
76 |
|
77 |
|
78 |
+
def _load_yolov5_model(self, pretrained: bool, model_version: str) -> None:
|
79 |
"""
|
80 |
Load the YOLOv5 model.
|
81 |
|
82 |
Args:
|
83 |
pretrained (bool): If True, load a pretrained model.
|
84 |
model_version (str): Version of the YOLOv5 model.
|
85 |
+
Raises:
|
86 |
+
Exception: If an error occurs during model loading.
|
87 |
"""
|
88 |
|
89 |
try:
|
|
|
97 |
raise
|
98 |
|
99 |
|
100 |
+
def process_image(self, image_input: Union[str, io.IOBase, Image.Image]) -> Image.Image:
|
101 |
"""
|
102 |
Process the image from the given path or file-like object.
|
103 |
+
|
104 |
Args:
|
105 |
+
image_input (Union[str, io.IOBase, Image.Image]): Path to the image file, a file-like object, or a PIL Image.
|
106 |
+
|
107 |
Returns:
|
108 |
Image.Image: Processed image in RGB format.
|
109 |
+
|
110 |
Raises:
|
111 |
Exception: If an error occurs during image processing.
|
112 |
"""
|
|
|
127 |
raise
|
128 |
|
129 |
|
130 |
+
def detect_objects(self, image: Image.Image, threshold: float = 0.4) -> Tuple[str, List[Tuple[str, List[float], float]]]:
|
131 |
"""
|
132 |
Detect objects in the given image using the loaded model.
|
133 |
|
134 |
Args:
|
135 |
image (Image.Image): Image in which to detect objects.
|
136 |
+
threshold (float): Model detection confidence threshold.
|
137 |
|
138 |
Returns:
|
139 |
+
Tuple[str, List[Tuple[str, List[float], float]]]: A tuple containing a string representation and a list of detected objects.
|
140 |
+
Each object in the list is represented as a tuple (label_name, box_rounded, certainty).
|
141 |
|
142 |
Raises:
|
143 |
ValueError: If the model is not loaded or the model name is unsupported.
|
|
|
151 |
raise ValueError("Model not loaded or unsupported model name")
|
152 |
|
153 |
|
154 |
+
def _detect_with_detic(self, image: Image.Image, threshold: float) -> Tuple[str, List[Tuple[str, List[float], float]]]:
|
155 |
"""
|
156 |
Detect objects using the Detic model.
|
157 |
|
|
|
160 |
threshold (float): The confidence threshold for detections.
|
161 |
|
162 |
Returns:
|
163 |
+
Tuple[str, List[Tuple[str, List[float], float]]]: A tuple containing a string representation and a list of detected objects.
|
164 |
+
Each object in the list is represented as a tuple (label_name, box_rounded, certainty).
|
165 |
"""
|
166 |
|
167 |
inputs = self.processor(images=image, return_tensors="pt")
|
|
|
180 |
return detected_objects_str, detected_objects_list
|
181 |
|
182 |
|
183 |
+
def _detect_with_yolov5(self, image: Image.Image, threshold: float) -> Tuple[str, List[Tuple[str, List[float], float]]]:
|
184 |
"""
|
185 |
Detect objects using the YOLOv5 model.
|
186 |
|
|
|
189 |
threshold (float): The confidence threshold for detections.
|
190 |
|
191 |
Returns:
|
192 |
+
Tuple[str, List[Tuple[str, List[float], float]]]: A tuple containing a string representation and a list of detected objects.
|
193 |
+
Each object in the list is represented as a tuple (label_name, box_rounded, certainty).
|
194 |
"""
|
195 |
|
196 |
cv2_img = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
|
|
|
207 |
return detected_objects_str, detected_objects_list
|
208 |
|
209 |
|
210 |
+
def draw_boxes(self, image: Image.Image, detected_objects: List[Tuple[str, List[float], float]], show_confidence: bool = True) -> Image.Image:
|
211 |
"""
|
212 |
Draw bounding boxes around detected objects in the image.
|
213 |
|
214 |
Args:
|
215 |
image (Image.Image): Image on which to draw.
|
216 |
+
detected_objects (List[Tuple[str, List[float], float]]): List of detected objects.
|
217 |
show_confidence (bool): Whether to show confidence scores.
|
218 |
|
219 |
Returns:
|
|
|
241 |
return image
|
242 |
|
243 |
|
244 |
+
def detect_and_draw_objects(image_path: str, model_type: str = 'yolov5', threshold: float = 0.2, show_confidence: bool = True) -> Tuple[Image.Image, str]:
|
245 |
"""
|
246 |
Detects objects in an image, draws bounding boxes around them, and returns the processed image and a string description.
|
247 |
|
|
|
252 |
show_confidence (bool): Whether to show confidence scores on the output image.
|
253 |
|
254 |
Returns:
|
255 |
+
Tuple[Image.Image, str]: A tuple containing the processed Image.Image and a string of detected objects.
|
256 |
"""
|
257 |
|
258 |
detector = ObjectDetector()
|
|
|
261 |
detected_objects_string, detected_objects_list = detector.detect_objects(image, threshold=threshold)
|
262 |
image_with_boxes = detector.draw_boxes(image, detected_objects_list, show_confidence=show_confidence)
|
263 |
return image_with_boxes, detected_objects_string
|
|
|
|
|
|
|
|
|
|
|
|