Hugging Face's logo

Spaces:
Sijuade
/
YOLOV3-GradCAM
Runtime error

App Files Community
Sijuade commited on
Commit
d91bd79
·
1 Parent(s): bfda67b

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +384 -1
README.md CHANGED
@@ -10,4 +10,387 @@ pinned: false
10
  license: mit
11
  ---
12
 
13
- Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
10
  license: mit
11
  ---
12
 
13
+ # Gradio Object Detection App with GradCAM for YOLOv3 - ERAv1 Session 13
14
+
15
+ ## Table of Contents
16
+ - [Introduction](#introduction)
17
+ - [Features](#features)
18
+ - [Model Performance](#model-performance)
19
+ - [Inference Samples](#inference-samples)
20
+ - [How to Use](#how-to-use)
21
+ - [Supported Classes](#supported-classes)
22
+ - [Link to the Model](#link-to-the-model)
23
+ - [Acknowledgements](#acknowledgements)
24
+
25
+ ## Introduction
26
+ This Gradio app showcases an object detection model using YOLOv3 architecture. The model is trained with enhanced features like multi-resolution training and Mosaic Augmentation. Additionally, the app provides GradCAM outputs for better visualization of the model's predictions.
27
+
28
+ ## Features
29
+ - **PytorchLightning Implementation**: The codebase has been refactored to use PytorchLightning for a more modular and scalable approach.
30
+ - **Multi-resolution Training**: Unlike traditional models that train on a fixed resolution, this model has been trained on multiple resolutions (416, 608, 896, 1280) for better generalization.
31
+ - **Mosaic Augmentation**: Implemented Mosaic Augmentation to enhance the training dataset, but only applied 75% of the time to maintain variety.
32
+ - **Precision Training**: The model is trained using float16 precision for faster convergence and reduced memory usage.
33
+ - **GradCAM Visualization**: Integrated GradCAM to provide a heatmap visualization of the regions in the image that the model focuses on during prediction.
34
+
35
+ ## Model Performance
36
+ ```
37
+ ┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
38
+ ┃ Validate metric ┃ DataLoader 0 ┃
39
+ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩
40
+ │ val_class_accuracy_epoch │ 81.89761352539062 │
41
+ │ val_loss │ 6.100630283355713 │
42
+ │ val_no_obj_accuracy_epoch │ 97.92534637451172 │
43
+ │ val_obj_accuracy_epoch │ 71.2684097290039 │
44
+ └───────────────────────────┴───────────────────────────┘
45
+ 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 619/619 [29:42<00:00, 2.88s/it]
46
+ MAP: 0.10860311985015869
47
+ ```
48
+ ## Inference Samples
49
+ ![pred1](https://github.com/Delve-ERAV1/S13/assets/11761529/df995d26-8d1b-44cd-8979-df4fd514ed44)
50
+ ![pred2](https://github.com/Delve-ERAV1/S13/assets/11761529/c343787c-1d39-44f6-86f5-c8c228e193e8)
51
+
52
+ ## How to Use
53
+ 1. Navigate to the Gradio app interface.
54
+ 2. Upload a custom image or select from the provided samples.
55
+ 3. Click on the "Predict" button.
56
+ 4. View the object detection predictions along with the GradCAM heatmap.
57
+
58
+ ## Supported Classes
59
+ ![supported_classes](https://github.com/Delve-ERAV1/S13/assets/11761529/49ef1748-9eed-4cca-b8d6-24200400bdf0)
60
+
61
+ ## Model Architecture
62
+ ```
63
+ ----------------------------------------------------------------
64
+ Layer (type) Output Shape Param #
65
+ ================================================================
66
+ Conv2d-1 [-1, 32, 416, 416] 864
67
+ BatchNorm2d-2 [-1, 32, 416, 416] 64
68
+ LeakyReLU-3 [-1, 32, 416, 416] 0
69
+ CNNBlock-4 [-1, 32, 416, 416] 0
70
+ Conv2d-5 [-1, 64, 208, 208] 18,432
71
+ BatchNorm2d-6 [-1, 64, 208, 208] 128
72
+ LeakyReLU-7 [-1, 64, 208, 208] 0
73
+ CNNBlock-8 [-1, 64, 208, 208] 0
74
+ Conv2d-9 [-1, 32, 208, 208] 2,048
75
+ BatchNorm2d-10 [-1, 32, 208, 208] 64
76
+ LeakyReLU-11 [-1, 32, 208, 208] 0
77
+ CNNBlock-12 [-1, 32, 208, 208] 0
78
+ Conv2d-13 [-1, 64, 208, 208] 18,432
79
+ BatchNorm2d-14 [-1, 64, 208, 208] 128
80
+ LeakyReLU-15 [-1, 64, 208, 208] 0
81
+ CNNBlock-16 [-1, 64, 208, 208] 0
82
+ ResidualBlock-17 [-1, 64, 208, 208] 0
83
+ Conv2d-18 [-1, 128, 104, 104] 73,728
84
+ BatchNorm2d-19 [-1, 128, 104, 104] 256
85
+ LeakyReLU-20 [-1, 128, 104, 104] 0
86
+ CNNBlock-21 [-1, 128, 104, 104] 0
87
+ Conv2d-22 [-1, 64, 104, 104] 8,192
88
+ BatchNorm2d-23 [-1, 64, 104, 104] 128
89
+ LeakyReLU-24 [-1, 64, 104, 104] 0
90
+ CNNBlock-25 [-1, 64, 104, 104] 0
91
+ Conv2d-26 [-1, 128, 104, 104] 73,728
92
+ BatchNorm2d-27 [-1, 128, 104, 104] 256
93
+ LeakyReLU-28 [-1, 128, 104, 104] 0
94
+ CNNBlock-29 [-1, 128, 104, 104] 0
95
+ Conv2d-30 [-1, 64, 104, 104] 8,192
96
+ BatchNorm2d-31 [-1, 64, 104, 104] 128
97
+ LeakyReLU-32 [-1, 64, 104, 104] 0
98
+ CNNBlock-33 [-1, 64, 104, 104] 0
99
+ Conv2d-34 [-1, 128, 104, 104] 73,728
100
+ BatchNorm2d-35 [-1, 128, 104, 104] 256
101
+ LeakyReLU-36 [-1, 128, 104, 104] 0
102
+ CNNBlock-37 [-1, 128, 104, 104] 0
103
+ ResidualBlock-38 [-1, 128, 104, 104] 0
104
+ Conv2d-39 [-1, 256, 52, 52] 294,912
105
+ BatchNorm2d-40 [-1, 256, 52, 52] 512
106
+ LeakyReLU-41 [-1, 256, 52, 52] 0
107
+ CNNBlock-42 [-1, 256, 52, 52] 0
108
+ Conv2d-43 [-1, 128, 52, 52] 32,768
109
+ BatchNorm2d-44 [-1, 128, 52, 52] 256
110
+ LeakyReLU-45 [-1, 128, 52, 52] 0
111
+ CNNBlock-46 [-1, 128, 52, 52] 0
112
+ Conv2d-47 [-1, 256, 52, 52] 294,912
113
+ BatchNorm2d-48 [-1, 256, 52, 52] 512
114
+ LeakyReLU-49 [-1, 256, 52, 52] 0
115
+ CNNBlock-50 [-1, 256, 52, 52] 0
116
+ Conv2d-51 [-1, 128, 52, 52] 32,768
117
+ BatchNorm2d-52 [-1, 128, 52, 52] 256
118
+ LeakyReLU-53 [-1, 128, 52, 52] 0
119
+ CNNBlock-54 [-1, 128, 52, 52] 0
120
+ Conv2d-55 [-1, 256, 52, 52] 294,912
121
+ BatchNorm2d-56 [-1, 256, 52, 52] 512
122
+ LeakyReLU-57 [-1, 256, 52, 52] 0
123
+ CNNBlock-58 [-1, 256, 52, 52] 0
124
+ Conv2d-59 [-1, 128, 52, 52] 32,768
125
+ BatchNorm2d-60 [-1, 128, 52, 52] 256
126
+ LeakyReLU-61 [-1, 128, 52, 52] 0
127
+ CNNBlock-62 [-1, 128, 52, 52] 0
128
+ Conv2d-63 [-1, 256, 52, 52] 294,912
129
+ BatchNorm2d-64 [-1, 256, 52, 52] 512
130
+ LeakyReLU-65 [-1, 256, 52, 52] 0
131
+ CNNBlock-66 [-1, 256, 52, 52] 0
132
+ Conv2d-67 [-1, 128, 52, 52] 32,768
133
+ BatchNorm2d-68 [-1, 128, 52, 52] 256
134
+ LeakyReLU-69 [-1, 128, 52, 52] 0
135
+ CNNBlock-70 [-1, 128, 52, 52] 0
136
+ Conv2d-71 [-1, 256, 52, 52] 294,912
137
+ BatchNorm2d-72 [-1, 256, 52, 52] 512
138
+ LeakyReLU-73 [-1, 256, 52, 52] 0
139
+ CNNBlock-74 [-1, 256, 52, 52] 0
140
+ Conv2d-75 [-1, 128, 52, 52] 32,768
141
+ BatchNorm2d-76 [-1, 128, 52, 52] 256
142
+ LeakyReLU-77 [-1, 128, 52, 52] 0
143
+ CNNBlock-78 [-1, 128, 52, 52] 0
144
+ Conv2d-79 [-1, 256, 52, 52] 294,912
145
+ BatchNorm2d-80 [-1, 256, 52, 52] 512
146
+ LeakyReLU-81 [-1, 256, 52, 52] 0
147
+ CNNBlock-82 [-1, 256, 52, 52] 0
148
+ Conv2d-83 [-1, 128, 52, 52] 32,768
149
+ BatchNorm2d-84 [-1, 128, 52, 52] 256
150
+ LeakyReLU-85 [-1, 128, 52, 52] 0
151
+ CNNBlock-86 [-1, 128, 52, 52] 0
152
+ Conv2d-87 [-1, 256, 52, 52] 294,912
153
+ BatchNorm2d-88 [-1, 256, 52, 52] 512
154
+ LeakyReLU-89 [-1, 256, 52, 52] 0
155
+ CNNBlock-90 [-1, 256, 52, 52] 0
156
+ Conv2d-91 [-1, 128, 52, 52] 32,768
157
+ BatchNorm2d-92 [-1, 128, 52, 52] 256
158
+ LeakyReLU-93 [-1, 128, 52, 52] 0
159
+ CNNBlock-94 [-1, 128, 52, 52] 0
160
+ Conv2d-95 [-1, 256, 52, 52] 294,912
161
+ BatchNorm2d-96 [-1, 256, 52, 52] 512
162
+ LeakyReLU-97 [-1, 256, 52, 52] 0
163
+ CNNBlock-98 [-1, 256, 52, 52] 0
164
+ Conv2d-99 [-1, 128, 52, 52] 32,768
165
+ BatchNorm2d-100 [-1, 128, 52, 52] 256
166
+ LeakyReLU-101 [-1, 128, 52, 52] 0
167
+ CNNBlock-102 [-1, 128, 52, 52] 0
168
+ Conv2d-103 [-1, 256, 52, 52] 294,912
169
+ BatchNorm2d-104 [-1, 256, 52, 52] 512
170
+ LeakyReLU-105 [-1, 256, 52, 52] 0
171
+ CNNBlock-106 [-1, 256, 52, 52] 0
172
+ ResidualBlock-107 [-1, 256, 52, 52] 0
173
+ Conv2d-108 [-1, 512, 26, 26] 1,179,648
174
+ BatchNorm2d-109 [-1, 512, 26, 26] 1,024
175
+ LeakyReLU-110 [-1, 512, 26, 26] 0
176
+ CNNBlock-111 [-1, 512, 26, 26] 0
177
+ Conv2d-112 [-1, 256, 26, 26] 131,072
178
+ BatchNorm2d-113 [-1, 256, 26, 26] 512
179
+ LeakyReLU-114 [-1, 256, 26, 26] 0
180
+ CNNBlock-115 [-1, 256, 26, 26] 0
181
+ Conv2d-116 [-1, 512, 26, 26] 1,179,648
182
+ BatchNorm2d-117 [-1, 512, 26, 26] 1,024
183
+ LeakyReLU-118 [-1, 512, 26, 26] 0
184
+ CNNBlock-119 [-1, 512, 26, 26] 0
185
+ Conv2d-120 [-1, 256, 26, 26] 131,072
186
+ BatchNorm2d-121 [-1, 256, 26, 26] 512
187
+ LeakyReLU-122 [-1, 256, 26, 26] 0
188
+ CNNBlock-123 [-1, 256, 26, 26] 0
189
+ Conv2d-124 [-1, 512, 26, 26] 1,179,648
190
+ BatchNorm2d-125 [-1, 512, 26, 26] 1,024
191
+ LeakyReLU-126 [-1, 512, 26, 26] 0
192
+ CNNBlock-127 [-1, 512, 26, 26] 0
193
+ Conv2d-128 [-1, 256, 26, 26] 131,072
194
+ BatchNorm2d-129 [-1, 256, 26, 26] 512
195
+ LeakyReLU-130 [-1, 256, 26, 26] 0
196
+ CNNBlock-131 [-1, 256, 26, 26] 0
197
+ Conv2d-132 [-1, 512, 26, 26] 1,179,648
198
+ BatchNorm2d-133 [-1, 512, 26, 26] 1,024
199
+ LeakyReLU-134 [-1, 512, 26, 26] 0
200
+ CNNBlock-135 [-1, 512, 26, 26] 0
201
+ Conv2d-136 [-1, 256, 26, 26] 131,072
202
+ BatchNorm2d-137 [-1, 256, 26, 26] 512
203
+ LeakyReLU-138 [-1, 256, 26, 26] 0
204
+ CNNBlock-139 [-1, 256, 26, 26] 0
205
+ Conv2d-140 [-1, 512, 26, 26] 1,179,648
206
+ BatchNorm2d-141 [-1, 512, 26, 26] 1,024
207
+ LeakyReLU-142 [-1, 512, 26, 26] 0
208
+ CNNBlock-143 [-1, 512, 26, 26] 0
209
+ Conv2d-144 [-1, 256, 26, 26] 131,072
210
+ BatchNorm2d-145 [-1, 256, 26, 26] 512
211
+ LeakyReLU-146 [-1, 256, 26, 26] 0
212
+ CNNBlock-147 [-1, 256, 26, 26] 0
213
+ Conv2d-148 [-1, 512, 26, 26] 1,179,648
214
+ BatchNorm2d-149 [-1, 512, 26, 26] 1,024
215
+ LeakyReLU-150 [-1, 512, 26, 26] 0
216
+ CNNBlock-151 [-1, 512, 26, 26] 0
217
+ Conv2d-152 [-1, 256, 26, 26] 131,072
218
+ BatchNorm2d-153 [-1, 256, 26, 26] 512
219
+ LeakyReLU-154 [-1, 256, 26, 26] 0
220
+ CNNBlock-155 [-1, 256, 26, 26] 0
221
+ Conv2d-156 [-1, 512, 26, 26] 1,179,648
222
+ BatchNorm2d-157 [-1, 512, 26, 26] 1,024
223
+ LeakyReLU-158 [-1, 512, 26, 26] 0
224
+ CNNBlock-159 [-1, 512, 26, 26] 0
225
+ Conv2d-160 [-1, 256, 26, 26] 131,072
226
+ BatchNorm2d-161 [-1, 256, 26, 26] 512
227
+ LeakyReLU-162 [-1, 256, 26, 26] 0
228
+ CNNBlock-163 [-1, 256, 26, 26] 0
229
+ Conv2d-164 [-1, 512, 26, 26] 1,179,648
230
+ BatchNorm2d-165 [-1, 512, 26, 26] 1,024
231
+ LeakyReLU-166 [-1, 512, 26, 26] 0
232
+ CNNBlock-167 [-1, 512, 26, 26] 0
233
+ Conv2d-168 [-1, 256, 26, 26] 131,072
234
+ BatchNorm2d-169 [-1, 256, 26, 26] 512
235
+ LeakyReLU-170 [-1, 256, 26, 26] 0
236
+ CNNBlock-171 [-1, 256, 26, 26] 0
237
+ Conv2d-172 [-1, 512, 26, 26] 1,179,648
238
+ BatchNorm2d-173 [-1, 512, 26, 26] 1,024
239
+ LeakyReLU-174 [-1, 512, 26, 26] 0
240
+ CNNBlock-175 [-1, 512, 26, 26] 0
241
+ ResidualBlock-176 [-1, 512, 26, 26] 0
242
+ Conv2d-177 [-1, 1024, 13, 13] 4,718,592
243
+ BatchNorm2d-178 [-1, 1024, 13, 13] 2,048
244
+ LeakyReLU-179 [-1, 1024, 13, 13] 0
245
+ CNNBlock-180 [-1, 1024, 13, 13] 0
246
+ Conv2d-181 [-1, 512, 13, 13] 524,288
247
+ BatchNorm2d-182 [-1, 512, 13, 13] 1,024
248
+ LeakyReLU-183 [-1, 512, 13, 13] 0
249
+ CNNBlock-184 [-1, 512, 13, 13] 0
250
+ Conv2d-185 [-1, 1024, 13, 13] 4,718,592
251
+ BatchNorm2d-186 [-1, 1024, 13, 13] 2,048
252
+ LeakyReLU-187 [-1, 1024, 13, 13] 0
253
+ CNNBlock-188 [-1, 1024, 13, 13] 0
254
+ Conv2d-189 [-1, 512, 13, 13] 524,288
255
+ BatchNorm2d-190 [-1, 512, 13, 13] 1,024
256
+ LeakyReLU-191 [-1, 512, 13, 13] 0
257
+ CNNBlock-192 [-1, 512, 13, 13] 0
258
+ Conv2d-193 [-1, 1024, 13, 13] 4,718,592
259
+ BatchNorm2d-194 [-1, 1024, 13, 13] 2,048
260
+ LeakyReLU-195 [-1, 1024, 13, 13] 0
261
+ CNNBlock-196 [-1, 1024, 13, 13] 0
262
+ Conv2d-197 [-1, 512, 13, 13] 524,288
263
+ BatchNorm2d-198 [-1, 512, 13, 13] 1,024
264
+ LeakyReLU-199 [-1, 512, 13, 13] 0
265
+ CNNBlock-200 [-1, 512, 13, 13] 0
266
+ Conv2d-201 [-1, 1024, 13, 13] 4,718,592
267
+ BatchNorm2d-202 [-1, 1024, 13, 13] 2,048
268
+ LeakyReLU-203 [-1, 1024, 13, 13] 0
269
+ CNNBlock-204 [-1, 1024, 13, 13] 0
270
+ Conv2d-205 [-1, 512, 13, 13] 524,288
271
+ BatchNorm2d-206 [-1, 512, 13, 13] 1,024
272
+ LeakyReLU-207 [-1, 512, 13, 13] 0
273
+ CNNBlock-208 [-1, 512, 13, 13] 0
274
+ Conv2d-209 [-1, 1024, 13, 13] 4,718,592
275
+ BatchNorm2d-210 [-1, 1024, 13, 13] 2,048
276
+ LeakyReLU-211 [-1, 1024, 13, 13] 0
277
+ CNNBlock-212 [-1, 1024, 13, 13] 0
278
+ ResidualBlock-213 [-1, 1024, 13, 13] 0
279
+ Conv2d-214 [-1, 1024, 13, 13] 1,048,576
280
+ BatchNorm2d-215 [-1, 1024, 13, 13] 2,048
281
+ LeakyReLU-216 [-1, 1024, 13, 13] 0
282
+ CNNBlock-217 [-1, 1024, 13, 13] 0
283
+ Conv2d-218 [-1, 2048, 13, 13] 18,874,368
284
+ BatchNorm2d-219 [-1, 2048, 13, 13] 4,096
285
+ LeakyReLU-220 [-1, 2048, 13, 13] 0
286
+ CNNBlock-221 [-1, 2048, 13, 13] 0
287
+ Conv2d-222 [-1, 1024, 13, 13] 2,097,152
288
+ BatchNorm2d-223 [-1, 1024, 13, 13] 2,048
289
+ LeakyReLU-224 [-1, 1024, 13, 13] 0
290
+ CNNBlock-225 [-1, 1024, 13, 13] 0
291
+ Conv2d-226 [-1, 2048, 13, 13] 18,874,368
292
+ BatchNorm2d-227 [-1, 2048, 13, 13] 4,096
293
+ LeakyReLU-228 [-1, 2048, 13, 13] 0
294
+ CNNBlock-229 [-1, 2048, 13, 13] 0
295
+ ResidualBlock-230 [-1, 2048, 13, 13] 0
296
+ Conv2d-231 [-1, 1024, 13, 13] 2,097,152
297
+ BatchNorm2d-232 [-1, 1024, 13, 13] 2,048
298
+ LeakyReLU-233 [-1, 1024, 13, 13] 0
299
+ CNNBlock-234 [-1, 1024, 13, 13] 0
300
+ Conv2d-235 [-1, 2048, 13, 13] 18,874,368
301
+ BatchNorm2d-236 [-1, 2048, 13, 13] 4,096
302
+ LeakyReLU-237 [-1, 2048, 13, 13] 0
303
+ CNNBlock-238 [-1, 2048, 13, 13] 0
304
+ Conv2d-239 [-1, 75, 13, 13] 153,675
305
+ CNNBlock-240 [-1, 75, 13, 13] 0
306
+ ScalePrediction-241 [-1, 3, 13, 13, 25] 0
307
+ Conv2d-242 [-1, 256, 13, 13] 262,144
308
+ BatchNorm2d-243 [-1, 256, 13, 13] 512
309
+ LeakyReLU-244 [-1, 256, 13, 13] 0
310
+ CNNBlock-245 [-1, 256, 13, 13] 0
311
+ Upsample-246 [-1, 256, 26, 26] 0
312
+ Conv2d-247 [-1, 256, 26, 26] 196,608
313
+ BatchNorm2d-248 [-1, 256, 26, 26] 512
314
+ LeakyReLU-249 [-1, 256, 26, 26] 0
315
+ CNNBlock-250 [-1, 256, 26, 26] 0
316
+ Conv2d-251 [-1, 512, 26, 26] 1,179,648
317
+ BatchNorm2d-252 [-1, 512, 26, 26] 1,024
318
+ LeakyReLU-253 [-1, 512, 26, 26] 0
319
+ CNNBlock-254 [-1, 512, 26, 26] 0
320
+ Conv2d-255 [-1, 256, 26, 26] 131,072
321
+ BatchNorm2d-256 [-1, 256, 26, 26] 512
322
+ LeakyReLU-257 [-1, 256, 26, 26] 0
323
+ CNNBlock-258 [-1, 256, 26, 26] 0
324
+ Conv2d-259 [-1, 512, 26, 26] 1,179,648
325
+ BatchNorm2d-260 [-1, 512, 26, 26] 1,024
326
+ LeakyReLU-261 [-1, 512, 26, 26] 0
327
+ CNNBlock-262 [-1, 512, 26, 26] 0
328
+ ResidualBlock-263 [-1, 512, 26, 26] 0
329
+ Conv2d-264 [-1, 256, 26, 26] 131,072
330
+ BatchNorm2d-265 [-1, 256, 26, 26] 512
331
+ LeakyReLU-266 [-1, 256, 26, 26] 0
332
+ CNNBlock-267 [-1, 256, 26, 26] 0
333
+ Conv2d-268 [-1, 512, 26, 26] 1,179,648
334
+ BatchNorm2d-269 [-1, 512, 26, 26] 1,024
335
+ LeakyReLU-270 [-1, 512, 26, 26] 0
336
+ CNNBlock-271 [-1, 512, 26, 26] 0
337
+ Conv2d-272 [-1, 75, 26, 26] 38,475
338
+ CNNBlock-273 [-1, 75, 26, 26] 0
339
+ ScalePrediction-274 [-1, 3, 26, 26, 25] 0
340
+ Conv2d-275 [-1, 128, 26, 26] 32,768
341
+ BatchNorm2d-276 [-1, 128, 26, 26] 256
342
+ LeakyReLU-277 [-1, 128, 26, 26] 0
343
+ CNNBlock-278 [-1, 128, 26, 26] 0
344
+ Upsample-279 [-1, 128, 52, 52] 0
345
+ Conv2d-280 [-1, 128, 52, 52] 49,152
346
+ BatchNorm2d-281 [-1, 128, 52, 52] 256
347
+ LeakyReLU-282 [-1, 128, 52, 52] 0
348
+ CNNBlock-283 [-1, 128, 52, 52] 0
349
+ Conv2d-284 [-1, 256, 52, 52] 294,912
350
+ BatchNorm2d-285 [-1, 256, 52, 52] 512
351
+ LeakyReLU-286 [-1, 256, 52, 52] 0
352
+ CNNBlock-287 [-1, 256, 52, 52] 0
353
+ Conv2d-288 [-1, 128, 52, 52] 32,768
354
+ BatchNorm2d-289 [-1, 128, 52, 52] 256
355
+ LeakyReLU-290 [-1, 128, 52, 52] 0
356
+ CNNBlock-291 [-1, 128, 52, 52] 0
357
+ Conv2d-292 [-1, 256, 52, 52] 294,912
358
+ BatchNorm2d-293 [-1, 256, 52, 52] 512
359
+ LeakyReLU-294 [-1, 256, 52, 52] 0
360
+ CNNBlock-295 [-1, 256, 52, 52] 0
361
+ ResidualBlock-296 [-1, 256, 52, 52] 0
362
+ Conv2d-297 [-1, 128, 52, 52] 32,768
363
+ BatchNorm2d-298 [-1, 128, 52, 52] 256
364
+ LeakyReLU-299 [-1, 128, 52, 52] 0
365
+ CNNBlock-300 [-1, 128, 52, 52] 0
366
+ Conv2d-301 [-1, 256, 52, 52] 294,912
367
+ BatchNorm2d-302 [-1, 256, 52, 52] 512
368
+ LeakyReLU-303 [-1, 256, 52, 52] 0
369
+ CNNBlock-304 [-1, 256, 52, 52] 0
370
+ Conv2d-305 [-1, 75, 52, 52] 19,275
371
+ CNNBlock-306 [-1, 75, 52, 52] 0
372
+ ScalePrediction-307 [-1, 3, 52, 52, 25] 0
373
+ ================================================================
374
+ Total params: 107,980,481
375
+ Trainable params: 107,980,481
376
+ Non-trainable params: 0
377
+ ----------------------------------------------------------------
378
+ Input size (MB): 1.98
379
+ Forward/backward pass size (MB): 1253.79
380
+ Params size (MB): 411.91
381
+ Estimated Total Size (MB): 1667.68
382
+ ----------------------------------------------------------------
383
+ ```
384
+
385
+ ## Examples
386
+ App includes some examples images for testing
387
+ ![examples_yolo](https://github.com/Delve-ERAV1/S13/assets/11761529/ca81abde-8193-4d3b-b7d3-989b47d2cc5f)
388
+
389
+ ## Github
390
+ Training code may be found [here](https://github.com/Delve-ERAV1/S13)
391
+
392
+ ## References
393
+ https://arxiv.org/abs/1804.02767
394
+ https://www.youtube.com/watch?v=Grir6TZbc1M
395
+ https://github.com/jacobgil/pytorch-grad-cam
396
+