|
|
--- |
|
|
title: ERA SESSION13 |
|
|
emoji: 🔥 |
|
|
colorFrom: indigo |
|
|
colorTo: indigo |
|
|
sdk: gradio |
|
|
sdk_version: 3.40.1 |
|
|
app_file: app.py |
|
|
pinned: false |
|
|
license: mit |
|
|
--- |
|
|
|
|
|
# ERA-SESSION13 YoloV3 with Pytorch Lightning & Gradio |
|
|
|
|
|
HF Link: https://huggingface.co/spaces/Navyabhat/Session13 |
|
|
|
|
|
### Achieved: |
|
|
1. **Training Loss: 3.680** |
|
|
2. **Validation Loss: 4.940** |
|
|
3. **Class accuracy: 81.601883%** |
|
|
4. **No obj accuracy: 97.991463%** |
|
|
5. **Obj accuracy: 75.976616%** |
|
|
6. **MAP: 0.4366795** |
|
|
|
|
|
### Tasks: |
|
|
1. :heavy_check_mark: Move the code to PytorchLightning |
|
|
2. :heavy_check_mark: Train the model to reach such that all of these are true: |
|
|
- Class accuracy is more than 75% |
|
|
- No Obj accuracy of more than 95% |
|
|
- Object Accuracy of more than 70% (assuming you had to reduce the kernel numbers, else 80/98/78) |
|
|
- Ideally trained till 40 epochs |
|
|
3. :heavy_check_mark: Add these training features: |
|
|
- Add multi-resolution training - the code shared trains only on one resolution 416 |
|
|
- Add Implement Mosaic Augmentation only 75% of the times |
|
|
- Train on float16 |
|
|
- GradCam must be implemented. |
|
|
4. :heavy_check_mark: Things that are allowed due to HW constraints: |
|
|
- Change of batch size |
|
|
- Change of resolution |
|
|
- Change of OCP parameters |
|
|
5. :heavy_check_mark: Once done: |
|
|
- Move the app to HuggingFace Spaces |
|
|
- Allow custom upload of images |
|
|
- Share some samples from the existing dataset |
|
|
- Show the GradCAM output for the image that the user uploads as well as for the samples. |
|
|
6. :heavy_check_mark: Mention things like: |
|
|
- classes that your model support |
|
|
- link to the actual model |
|
|
7. :heavy_check_mark: Assignment: |
|
|
- Share HuggingFace App Link |
|
|
- Share LightningCode Link on Github |
|
|
- Share notebook link (with logs) on GitHub |
|
|
|
|
|
### Results |
|
|
 |
|
|
|
|
|
### Gradio App |
|
|
 |
|
|
 |
|
|
|
|
|
### Model Summary |
|
|
```python |
|
|
| Name | Type | Params |
|
|
------------------------------------------------------------------- |
|
|
0 | loss_fn | YoloLoss | 0 |
|
|
1 | loss_fn.mse | MSELoss | 0 |
|
|
2 | loss_fn.bce | BCEWithLogitsLoss | 0 |
|
|
3 | loss_fn.entropy | CrossEntropyLoss | 0 |
|
|
4 | loss_fn.sigmoid | Sigmoid | 0 |
|
|
5 | layers | ModuleList | 61.6 M |
|
|
6 | layers.0 | CNNBlock | 928 |
|
|
7 | layers.0.conv | Conv2d | 864 |
|
|
8 | layers.0.bn | BatchNorm2d | 64 |
|
|
9 | layers.0.leaky | LeakyReLU | 0 |
|
|
10 | layers.1 | CNNBlock | 18.6 K |
|
|
11 | layers.1.conv | Conv2d | 18.4 K |
|
|
12 | layers.1.bn | BatchNorm2d | 128 |
|
|
13 | layers.1.leaky | LeakyReLU | 0 |
|
|
14 | layers.2 | ResidualBlock | 20.7 K |
|
|
15 | layers.2.layers | ModuleList | 20.7 K |
|
|
16 | layers.2.layers.0 | Sequential | 20.7 K |
|
|
17 | layers.2.layers.0.0 | CNNBlock | 2.1 K |
|
|
18 | layers.2.layers.0.0.conv | Conv2d | 2.0 K |
|
|
19 | layers.2.layers.0.0.bn | BatchNorm2d | 64 |
|
|
20 | layers.2.layers.0.0.leaky | LeakyReLU | 0 |
|
|
21 | layers.2.layers.0.1 | CNNBlock | 18.6 K |
|
|
22 | layers.2.layers.0.1.conv | Conv2d | 18.4 K |
|
|
23 | layers.2.layers.0.1.bn | BatchNorm2d | 128 |
|
|
24 | layers.2.layers.0.1.leaky | LeakyReLU | 0 |
|
|
25 | layers.3 | CNNBlock | 74.0 K |
|
|
26 | layers.3.conv | Conv2d | 73.7 K |
|
|
27 | layers.3.bn | BatchNorm2d | 256 |
|
|
28 | layers.3.leaky | LeakyReLU | 0 |
|
|
29 | layers.4 | ResidualBlock | 164 K |
|
|
30 | layers.4.layers | ModuleList | 164 K |
|
|
31 | layers.4.layers.0 | Sequential | 82.3 K |
|
|
32 | layers.4.layers.0.0 | CNNBlock | 8.3 K |
|
|
33 | layers.4.layers.0.0.conv | Conv2d | 8.2 K |
|
|
34 | layers.4.layers.0.0.bn | BatchNorm2d | 128 |
|
|
35 | layers.4.layers.0.0.leaky | LeakyReLU | 0 |
|
|
36 | layers.4.layers.0.1 | CNNBlock | 74.0 K |
|
|
37 | layers.4.layers.0.1.conv | Conv2d | 73.7 K |
|
|
38 | layers.4.layers.0.1.bn | BatchNorm2d | 256 |
|
|
39 | layers.4.layers.0.1.leaky | LeakyReLU | 0 |
|
|
40 | layers.4.layers.1 | Sequential | 82.3 K |
|
|
41 | layers.4.layers.1.0 | CNNBlock | 8.3 K |
|
|
42 | layers.4.layers.1.0.conv | Conv2d | 8.2 K |
|
|
43 | layers.4.layers.1.0.bn | BatchNorm2d | 128 |
|
|
44 | layers.4.layers.1.0.leaky | LeakyReLU | 0 |
|
|
45 | layers.4.layers.1.1 | CNNBlock | 74.0 K |
|
|
46 | layers.4.layers.1.1.conv | Conv2d | 73.7 K |
|
|
47 | layers.4.layers.1.1.bn | BatchNorm2d | 256 |
|
|
48 | layers.4.layers.1.1.leaky | LeakyReLU | 0 |
|
|
49 | layers.5 | CNNBlock | 295 K |
|
|
50 | layers.5.conv | Conv2d | 294 K |
|
|
51 | layers.5.bn | BatchNorm2d | 512 |
|
|
52 | layers.5.leaky | LeakyReLU | 0 |
|
|
53 | layers.6 | ResidualBlock | 2.6 M |
|
|
54 | layers.6.layers | ModuleList | 2.6 M |
|
|
55 | layers.6.layers.0 | Sequential | 328 K |
|
|
56 | layers.6.layers.0.0 | CNNBlock | 33.0 K |
|
|
57 | layers.6.layers.0.0.conv | Conv2d | 32.8 K |
|
|
58 | layers.6.layers.0.0.bn | BatchNorm2d | 256 |
|
|
59 | layers.6.layers.0.0.leaky | LeakyReLU | 0 |
|
|
60 | layers.6.layers.0.1 | CNNBlock | 295 K |
|
|
61 | layers.6.layers.0.1.conv | Conv2d | 294 K |
|
|
62 | layers.6.layers.0.1.bn | BatchNorm2d | 512 |
|
|
63 | layers.6.layers.0.1.leaky | LeakyReLU | 0 |
|
|
64 | layers.6.layers.1 | Sequential | 328 K |
|
|
65 | layers.6.layers.1.0 | CNNBlock | 33.0 K |
|
|
66 | layers.6.layers.1.0.conv | Conv2d | 32.8 K |
|
|
67 | layers.6.layers.1.0.bn | BatchNorm2d | 256 |
|
|
68 | layers.6.layers.1.0.leaky | LeakyReLU | 0 |
|
|
69 | layers.6.layers.1.1 | CNNBlock | 295 K |
|
|
70 | layers.6.layers.1.1.conv | Conv2d | 294 K |
|
|
71 | layers.6.layers.1.1.bn | BatchNorm2d | 512 |
|
|
72 | layers.6.layers.1.1.leaky | LeakyReLU | 0 |
|
|
73 | layers.6.layers.2 | Sequential | 328 K |
|
|
74 | layers.6.layers.2.0 | CNNBlock | 33.0 K |
|
|
75 | layers.6.layers.2.0.conv | Conv2d | 32.8 K |
|
|
76 | layers.6.layers.2.0.bn | BatchNorm2d | 256 |
|
|
77 | layers.6.layers.2.0.leaky | LeakyReLU | 0 |
|
|
78 | layers.6.layers.2.1 | CNNBlock | 295 K |
|
|
79 | layers.6.layers.2.1.conv | Conv2d | 294 K |
|
|
80 | layers.6.layers.2.1.bn | BatchNorm2d | 512 |
|
|
81 | layers.6.layers.2.1.leaky | LeakyReLU | 0 |
|
|
82 | layers.6.layers.3 | Sequential | 328 K |
|
|
83 | layers.6.layers.3.0 | CNNBlock | 33.0 K |
|
|
84 | layers.6.layers.3.0.conv | Conv2d | 32.8 K |
|
|
85 | layers.6.layers.3.0.bn | BatchNorm2d | 256 |
|
|
86 | layers.6.layers.3.0.leaky | LeakyReLU | 0 |
|
|
87 | layers.6.layers.3.1 | CNNBlock | 295 K |
|
|
88 | layers.6.layers.3.1.conv | Conv2d | 294 K |
|
|
89 | layers.6.layers.3.1.bn | BatchNorm2d | 512 |
|
|
90 | layers.6.layers.3.1.leaky | LeakyReLU | 0 |
|
|
91 | layers.6.layers.4 | Sequential | 328 K |
|
|
92 | layers.6.layers.4.0 | CNNBlock | 33.0 K |
|
|
93 | layers.6.layers.4.0.conv | Conv2d | 32.8 K |
|
|
94 | layers.6.layers.4.0.bn | BatchNorm2d | 256 |
|
|
95 | layers.6.layers.4.0.leaky | LeakyReLU | 0 |
|
|
96 | layers.6.layers.4.1 | CNNBlock | 295 K |
|
|
97 | layers.6.layers.4.1.conv | Conv2d | 294 K |
|
|
98 | layers.6.layers.4.1.bn | BatchNorm2d | 512 |
|
|
99 | layers.6.layers.4.1.leaky | LeakyReLU | 0 |
|
|
100 | layers.6.layers.5 | Sequential | 328 K |
|
|
101 | layers.6.layers.5.0 | CNNBlock | 33.0 K |
|
|
102 | layers.6.layers.5.0.conv | Conv2d | 32.8 K |
|
|
103 | layers.6.layers.5.0.bn | BatchNorm2d | 256 |
|
|
104 | layers.6.layers.5.0.leaky | LeakyReLU | 0 |
|
|
105 | layers.6.layers.5.1 | CNNBlock | 295 K |
|
|
106 | layers.6.layers.5.1.conv | Conv2d | 294 K |
|
|
107 | layers.6.layers.5.1.bn | BatchNorm2d | 512 |
|
|
108 | layers.6.layers.5.1.leaky | LeakyReLU | 0 |
|
|
109 | layers.6.layers.6 | Sequential | 328 K |
|
|
110 | layers.6.layers.6.0 | CNNBlock | 33.0 K |
|
|
111 | layers.6.layers.6.0.conv | Conv2d | 32.8 K |
|
|
112 | layers.6.layers.6.0.bn | BatchNorm2d | 256 |
|
|
113 | layers.6.layers.6.0.leaky | LeakyReLU | 0 |
|
|
114 | layers.6.layers.6.1 | CNNBlock | 295 K |
|
|
115 | layers.6.layers.6.1.conv | Conv2d | 294 K |
|
|
116 | layers.6.layers.6.1.bn | BatchNorm2d | 512 |
|
|
117 | layers.6.layers.6.1.leaky | LeakyReLU | 0 |
|
|
118 | layers.6.layers.7 | Sequential | 328 K |
|
|
119 | layers.6.layers.7.0 | CNNBlock | 33.0 K |
|
|
120 | layers.6.layers.7.0.conv | Conv2d | 32.8 K |
|
|
121 | layers.6.layers.7.0.bn | BatchNorm2d | 256 |
|
|
122 | layers.6.layers.7.0.leaky | LeakyReLU | 0 |
|
|
123 | layers.6.layers.7.1 | CNNBlock | 295 K |
|
|
124 | layers.6.layers.7.1.conv | Conv2d | 294 K |
|
|
125 | layers.6.layers.7.1.bn | BatchNorm2d | 512 |
|
|
126 | layers.6.layers.7.1.leaky | LeakyReLU | 0 |
|
|
127 | layers.7 | CNNBlock | 1.2 M |
|
|
128 | layers.7.conv | Conv2d | 1.2 M |
|
|
129 | layers.7.bn | BatchNorm2d | 1.0 K |
|
|
130 | layers.7.leaky | LeakyReLU | 0 |
|
|
131 | layers.8 | ResidualBlock | 10.5 M |
|
|
132 | layers.8.layers | ModuleList | 10.5 M |
|
|
133 | layers.8.layers.0 | Sequential | 1.3 M |
|
|
134 | layers.8.layers.0.0 | CNNBlock | 131 K |
|
|
135 | layers.8.layers.0.0.conv | Conv2d | 131 K |
|
|
136 | layers.8.layers.0.0.bn | BatchNorm2d | 512 |
|
|
137 | layers.8.layers.0.0.leaky | LeakyReLU | 0 |
|
|
138 | layers.8.layers.0.1 | CNNBlock | 1.2 M |
|
|
139 | layers.8.layers.0.1.conv | Conv2d | 1.2 M |
|
|
140 | layers.8.layers.0.1.bn | BatchNorm2d | 1.0 K |
|
|
141 | layers.8.layers.0.1.leaky | LeakyReLU | 0 |
|
|
142 | layers.8.layers.1 | Sequential | 1.3 M |
|
|
143 | layers.8.layers.1.0 | CNNBlock | 131 K |
|
|
144 | layers.8.layers.1.0.conv | Conv2d | 131 K |
|
|
145 | layers.8.layers.1.0.bn | BatchNorm2d | 512 |
|
|
146 | layers.8.layers.1.0.leaky | LeakyReLU | 0 |
|
|
147 | layers.8.layers.1.1 | CNNBlock | 1.2 M |
|
|
148 | layers.8.layers.1.1.conv | Conv2d | 1.2 M |
|
|
149 | layers.8.layers.1.1.bn | BatchNorm2d | 1.0 K |
|
|
150 | layers.8.layers.1.1.leaky | LeakyReLU | 0 |
|
|
151 | layers.8.layers.2 | Sequential | 1.3 M |
|
|
152 | layers.8.layers.2.0 | CNNBlock | 131 K |
|
|
153 | layers.8.layers.2.0.conv | Conv2d | 131 K |
|
|
154 | layers.8.layers.2.0.bn | BatchNorm2d | 512 |
|
|
155 | layers.8.layers.2.0.leaky | LeakyReLU | 0 |
|
|
156 | layers.8.layers.2.1 | CNNBlock | 1.2 M |
|
|
157 | layers.8.layers.2.1.conv | Conv2d | 1.2 M |
|
|
158 | layers.8.layers.2.1.bn | BatchNorm2d | 1.0 K |
|
|
159 | layers.8.layers.2.1.leaky | LeakyReLU | 0 |
|
|
160 | layers.8.layers.3 | Sequential | 1.3 M |
|
|
161 | layers.8.layers.3.0 | CNNBlock | 131 K |
|
|
162 | layers.8.layers.3.0.conv | Conv2d | 131 K |
|
|
163 | layers.8.layers.3.0.bn | BatchNorm2d | 512 |
|
|
164 | layers.8.layers.3.0.leaky | LeakyReLU | 0 |
|
|
165 | layers.8.layers.3.1 | CNNBlock | 1.2 M |
|
|
166 | layers.8.layers.3.1.conv | Conv2d | 1.2 M |
|
|
167 | layers.8.layers.3.1.bn | BatchNorm2d | 1.0 K |
|
|
168 | layers.8.layers.3.1.leaky | LeakyReLU | 0 |
|
|
169 | layers.8.layers.4 | Sequential | 1.3 M |
|
|
170 | layers.8.layers.4.0 | CNNBlock | 131 K |
|
|
171 | layers.8.layers.4.0.conv | Conv2d | 131 K |
|
|
172 | layers.8.layers.4.0.bn | BatchNorm2d | 512 |
|
|
173 | layers.8.layers.4.0.leaky | LeakyReLU | 0 |
|
|
174 | layers.8.layers.4.1 | CNNBlock | 1.2 M |
|
|
175 | layers.8.layers.4.1.conv | Conv2d | 1.2 M |
|
|
176 | layers.8.layers.4.1.bn | BatchNorm2d | 1.0 K |
|
|
177 | layers.8.layers.4.1.leaky | LeakyReLU | 0 |
|
|
178 | layers.8.layers.5 | Sequential | 1.3 M |
|
|
179 | layers.8.layers.5.0 | CNNBlock | 131 K |
|
|
180 | layers.8.layers.5.0.conv | Conv2d | 131 K |
|
|
181 | layers.8.layers.5.0.bn | BatchNorm2d | 512 |
|
|
182 | layers.8.layers.5.0.leaky | LeakyReLU | 0 |
|
|
183 | layers.8.layers.5.1 | CNNBlock | 1.2 M |
|
|
184 | layers.8.layers.5.1.conv | Conv2d | 1.2 M |
|
|
185 | layers.8.layers.5.1.bn | BatchNorm2d | 1.0 K |
|
|
186 | layers.8.layers.5.1.leaky | LeakyReLU | 0 |
|
|
187 | layers.8.layers.6 | Sequential | 1.3 M |
|
|
188 | layers.8.layers.6.0 | CNNBlock | 131 K |
|
|
189 | layers.8.layers.6.0.conv | Conv2d | 131 K |
|
|
190 | layers.8.layers.6.0.bn | BatchNorm2d | 512 |
|
|
191 | layers.8.layers.6.0.leaky | LeakyReLU | 0 |
|
|
192 | layers.8.layers.6.1 | CNNBlock | 1.2 M |
|
|
193 | layers.8.layers.6.1.conv | Conv2d | 1.2 M |
|
|
194 | layers.8.layers.6.1.bn | BatchNorm2d | 1.0 K |
|
|
195 | layers.8.layers.6.1.leaky | LeakyReLU | 0 |
|
|
196 | layers.8.layers.7 | Sequential | 1.3 M |
|
|
197 | layers.8.layers.7.0 | CNNBlock | 131 K |
|
|
198 | layers.8.layers.7.0.conv | Conv2d | 131 K |
|
|
199 | layers.8.layers.7.0.bn | BatchNorm2d | 512 |
|
|
200 | layers.8.layers.7.0.leaky | LeakyReLU | 0 |
|
|
201 | layers.8.layers.7.1 | CNNBlock | 1.2 M |
|
|
202 | layers.8.layers.7.1.conv | Conv2d | 1.2 M |
|
|
203 | layers.8.layers.7.1.bn | BatchNorm2d | 1.0 K |
|
|
204 | layers.8.layers.7.1.leaky | LeakyReLU | 0 |
|
|
205 | layers.9 | CNNBlock | 4.7 M |
|
|
206 | layers.9.conv | Conv2d | 4.7 M |
|
|
207 | layers.9.bn | BatchNorm2d | 2.0 K |
|
|
208 | layers.9.leaky | LeakyReLU | 0 |
|
|
209 | layers.10 | ResidualBlock | 21.0 M |
|
|
210 | layers.10.layers | ModuleList | 21.0 M |
|
|
211 | layers.10.layers.0 | Sequential | 5.2 M |
|
|
212 | layers.10.layers.0.0 | CNNBlock | 525 K |
|
|
213 | layers.10.layers.0.0.conv | Conv2d | 524 K |
|
|
214 | layers.10.layers.0.0.bn | BatchNorm2d | 1.0 K |
|
|
215 | layers.10.layers.0.0.leaky | LeakyReLU | 0 |
|
|
216 | layers.10.layers.0.1 | CNNBlock | 4.7 M |
|
|
217 | layers.10.layers.0.1.conv | Conv2d | 4.7 M |
|
|
218 | layers.10.layers.0.1.bn | BatchNorm2d | 2.0 K |
|
|
219 | layers.10.layers.0.1.leaky | LeakyReLU | 0 |
|
|
220 | layers.10.layers.1 | Sequential | 5.2 M |
|
|
221 | layers.10.layers.1.0 | CNNBlock | 525 K |
|
|
222 | layers.10.layers.1.0.conv | Conv2d | 524 K |
|
|
223 | layers.10.layers.1.0.bn | BatchNorm2d | 1.0 K |
|
|
224 | layers.10.layers.1.0.leaky | LeakyReLU | 0 |
|
|
225 | layers.10.layers.1.1 | CNNBlock | 4.7 M |
|
|
226 | layers.10.layers.1.1.conv | Conv2d | 4.7 M |
|
|
227 | layers.10.layers.1.1.bn | BatchNorm2d | 2.0 K |
|
|
228 | layers.10.layers.1.1.leaky | LeakyReLU | 0 |
|
|
229 | layers.10.layers.2 | Sequential | 5.2 M |
|
|
230 | layers.10.layers.2.0 | CNNBlock | 525 K |
|
|
231 | layers.10.layers.2.0.conv | Conv2d | 524 K |
|
|
232 | layers.10.layers.2.0.bn | BatchNorm2d | 1.0 K |
|
|
233 | layers.10.layers.2.0.leaky | LeakyReLU | 0 |
|
|
234 | layers.10.layers.2.1 | CNNBlock | 4.7 M |
|
|
235 | layers.10.layers.2.1.conv | Conv2d | 4.7 M |
|
|
236 | layers.10.layers.2.1.bn | BatchNorm2d | 2.0 K |
|
|
237 | layers.10.layers.2.1.leaky | LeakyReLU | 0 |
|
|
238 | layers.10.layers.3 | Sequential | 5.2 M |
|
|
239 | layers.10.layers.3.0 | CNNBlock | 525 K |
|
|
240 | layers.10.layers.3.0.conv | Conv2d | 524 K |
|
|
241 | layers.10.layers.3.0.bn | BatchNorm2d | 1.0 K |
|
|
242 | layers.10.layers.3.0.leaky | LeakyReLU | 0 |
|
|
243 | layers.10.layers.3.1 | CNNBlock | 4.7 M |
|
|
244 | layers.10.layers.3.1.conv | Conv2d | 4.7 M |
|
|
245 | layers.10.layers.3.1.bn | BatchNorm2d | 2.0 K |
|
|
246 | layers.10.layers.3.1.leaky | LeakyReLU | 0 |
|
|
247 | layers.11 | CNNBlock | 525 K |
|
|
248 | layers.11.conv | Conv2d | 524 K |
|
|
249 | layers.11.bn | BatchNorm2d | 1.0 K |
|
|
250 | layers.11.leaky | LeakyReLU | 0 |
|
|
251 | layers.12 | CNNBlock | 4.7 M |
|
|
252 | layers.12.conv | Conv2d | 4.7 M |
|
|
253 | layers.12.bn | BatchNorm2d | 2.0 K |
|
|
254 | layers.12.leaky | LeakyReLU | 0 |
|
|
255 | layers.13 | ResidualBlock | 5.2 M |
|
|
256 | layers.13.layers | ModuleList | 5.2 M |
|
|
257 | layers.13.layers.0 | Sequential | 5.2 M |
|
|
258 | layers.13.layers.0.0 | CNNBlock | 525 K |
|
|
259 | layers.13.layers.0.0.conv | Conv2d | 524 K |
|
|
260 | layers.13.layers.0.0.bn | BatchNorm2d | 1.0 K |
|
|
261 | layers.13.layers.0.0.leaky | LeakyReLU | 0 |
|
|
262 | layers.13.layers.0.1 | CNNBlock | 4.7 M |
|
|
263 | layers.13.layers.0.1.conv | Conv2d | 4.7 M |
|
|
264 | layers.13.layers.0.1.bn | BatchNorm2d | 2.0 K |
|
|
265 | layers.13.layers.0.1.leaky | LeakyReLU | 0 |
|
|
266 | layers.14 | CNNBlock | 525 K |
|
|
267 | layers.14.conv | Conv2d | 524 K |
|
|
268 | layers.14.bn | BatchNorm2d | 1.0 K |
|
|
269 | layers.14.leaky | LeakyReLU | 0 |
|
|
270 | layers.15 | ScalePrediction | 4.8 M |
|
|
271 | layers.15.pred | Sequential | 4.8 M |
|
|
272 | layers.15.pred.0 | CNNBlock | 4.7 M |
|
|
273 | layers.15.pred.0.conv | Conv2d | 4.7 M |
|
|
274 | layers.15.pred.0.bn | BatchNorm2d | 2.0 K |
|
|
275 | layers.15.pred.0.leaky | LeakyReLU | 0 |
|
|
276 | layers.15.pred.1 | CNNBlock | 77.0 K |
|
|
277 | layers.15.pred.1.conv | Conv2d | 76.9 K |
|
|
278 | layers.15.pred.1.bn | BatchNorm2d | 150 |
|
|
279 | layers.15.pred.1.leaky | LeakyReLU | 0 |
|
|
280 | layers.16 | CNNBlock | 131 K |
|
|
281 | layers.16.conv | Conv2d | 131 K |
|
|
282 | layers.16.bn | BatchNorm2d | 512 |
|
|
283 | layers.16.leaky | LeakyReLU | 0 |
|
|
284 | layers.17 | Upsample | 0 |
|
|
285 | layers.18 | CNNBlock | 197 K |
|
|
286 | layers.18.conv | Conv2d | 196 K |
|
|
287 | layers.18.bn | BatchNorm2d | 512 |
|
|
288 | layers.18.leaky | LeakyReLU | 0 |
|
|
289 | layers.19 | CNNBlock | 1.2 M |
|
|
290 | layers.19.conv | Conv2d | 1.2 M |
|
|
291 | layers.19.bn | BatchNorm2d | 1.0 K |
|
|
292 | layers.19.leaky | LeakyReLU | 0 |
|
|
293 | layers.20 | ResidualBlock | 1.3 M |
|
|
294 | layers.20.layers | ModuleList | 1.3 M |
|
|
295 | layers.20.layers.0 | Sequential | 1.3 M |
|
|
296 | layers.20.layers.0.0 | CNNBlock | 131 K |
|
|
297 | layers.20.layers.0.0.conv | Conv2d | 131 K |
|
|
298 | layers.20.layers.0.0.bn | BatchNorm2d | 512 |
|
|
299 | layers.20.layers.0.0.leaky | LeakyReLU | 0 |
|
|
300 | layers.20.layers.0.1 | CNNBlock | 1.2 M |
|
|
301 | layers.20.layers.0.1.conv | Conv2d | 1.2 M |
|
|
302 | layers.20.layers.0.1.bn | BatchNorm2d | 1.0 K |
|
|
303 | layers.20.layers.0.1.leaky | LeakyReLU | 0 |
|
|
304 | layers.21 | CNNBlock | 131 K |
|
|
305 | layers.21.conv | Conv2d | 131 K |
|
|
306 | layers.21.bn | BatchNorm2d | 512 |
|
|
307 | layers.21.leaky | LeakyReLU | 0 |
|
|
308 | layers.22 | ScalePrediction | 1.2 M |
|
|
309 | layers.22.pred | Sequential | 1.2 M |
|
|
310 | layers.22.pred.0 | CNNBlock | 1.2 M |
|
|
311 | layers.22.pred.0.conv | Conv2d | 1.2 M |
|
|
312 | layers.22.pred.0.bn | BatchNorm2d | 1.0 K |
|
|
313 | layers.22.pred.0.leaky | LeakyReLU | 0 |
|
|
314 | layers.22.pred.1 | CNNBlock | 38.6 K |
|
|
315 | layers.22.pred.1.conv | Conv2d | 38.5 K |
|
|
316 | layers.22.pred.1.bn | BatchNorm2d | 150 |
|
|
317 | layers.22.pred.1.leaky | LeakyReLU | 0 |
|
|
318 | layers.23 | CNNBlock | 33.0 K |
|
|
319 | layers.23.conv | Conv2d | 32.8 K |
|
|
320 | layers.23.bn | BatchNorm2d | 256 |
|
|
321 | layers.23.leaky | LeakyReLU | 0 |
|
|
322 | layers.24 | Upsample | 0 |
|
|
323 | layers.25 | CNNBlock | 49.4 K |
|
|
324 | layers.25.conv | Conv2d | 49.2 K |
|
|
325 | layers.25.bn | BatchNorm2d | 256 |
|
|
326 | layers.25.leaky | LeakyReLU | 0 |
|
|
327 | layers.26 | CNNBlock | 295 K |
|
|
328 | layers.26.conv | Conv2d | 294 K |
|
|
329 | layers.26.bn | BatchNorm2d | 512 |
|
|
330 | layers.26.leaky | LeakyReLU | 0 |
|
|
331 | layers.27 | ResidualBlock | 328 K |
|
|
332 | layers.27.layers | ModuleList | 328 K |
|
|
333 | layers.27.layers.0 | Sequential | 328 K |
|
|
334 | layers.27.layers.0.0 | CNNBlock | 33.0 K |
|
|
335 | layers.27.layers.0.0.conv | Conv2d | 32.8 K |
|
|
336 | layers.27.layers.0.0.bn | BatchNorm2d | 256 |
|
|
337 | layers.27.layers.0.0.leaky | LeakyReLU | 0 |
|
|
338 | layers.27.layers.0.1 | CNNBlock | 295 K |
|
|
339 | layers.27.layers.0.1.conv | Conv2d | 294 K |
|
|
340 | layers.27.layers.0.1.bn | BatchNorm2d | 512 |
|
|
341 | layers.27.layers.0.1.leaky | LeakyReLU | 0 |
|
|
342 | layers.28 | CNNBlock | 33.0 K |
|
|
343 | layers.28.conv | Conv2d | 32.8 K |
|
|
344 | layers.28.bn | BatchNorm2d | 256 |
|
|
345 | layers.28.leaky | LeakyReLU | 0 |
|
|
346 | layers.29 | ScalePrediction | 314 K |
|
|
347 | layers.29.pred | Sequential | 314 K |
|
|
348 | layers.29.pred.0 | CNNBlock | 295 K |
|
|
349 | layers.29.pred.0.conv | Conv2d | 294 K |
|
|
350 | layers.29.pred.0.bn | BatchNorm2d | 512 |
|
|
351 | layers.29.pred.0.leaky | LeakyReLU | 0 |
|
|
352 | layers.29.pred.1 | CNNBlock | 19.4 K |
|
|
353 | layers.29.pred.1.conv | Conv2d | 19.3 K |
|
|
354 | layers.29.pred.1.bn | BatchNorm2d | 150 |
|
|
355 | layers.29.pred.1.leaky | LeakyReLU | 0 |
|
|
------------------------------------------------------------------- |
|
|
61.6 M Trainable params |
|
|
0 Non-trainable params |
|
|
61.6 M Total params |
|
|
246.506 Total estimated model params size (MB) |
|
|
``` |
|
|
|
|
|
### LR Finder |
|
|
 |
|
|
|
|
|
### Loss & Accuracy |
|
|
**Training & Validation Loss:** |
|
|
 |
|
|
|
|
|
**Testing Accuracy:** |
|
|
```python |
|
|
0%| | 0/39 [00:00<?, ?it/s] |
|
|
3%|▎ | 1/39 [00:05<03:24, 5.37s/it] |
|
|
5%|▌ | 2/39 [00:11<03:32, 5.75s/it] |
|
|
8%|▊ | 3/39 [00:16<03:14, 5.41s/it] |
|
|
10%|█ | 4/39 [00:21<03:06, 5.33s/it] |
|
|
13%|█▎ | 5/39 [00:26<02:55, 5.17s/it] |
|
|
15%|█▌ | 6/39 [00:31<02:50, 5.16s/it] |
|
|
18%|█▊ | 7/39 [00:36<02:43, 5.11s/it] |
|
|
21%|██ | 8/39 [00:42<02:48, 5.43s/it] |
|
|
23%|██▎ | 9/39 [00:48<02:44, 5.47s/it] |
|
|
26%|██▌ | 10/39 [00:54<02:41, 5.58s/it] |
|
|
28%|██▊ | 11/39 [00:59<02:36, 5.59s/it] |
|
|
31%|███ | 12/39 [01:05<02:35, 5.77s/it] |
|
|
33%|███▎ | 13/39 [01:11<02:28, 5.70s/it] |
|
|
36%|███▌ | 14/39 [01:16<02:15, 5.42s/it] |
|
|
38%|███▊ | 15/39 [01:21<02:07, 5.30s/it] |
|
|
41%|████ | 16/39 [01:26<02:02, 5.34s/it] |
|
|
44%|████▎ | 17/39 [01:31<01:54, 5.23s/it] |
|
|
46%|████▌ | 18/39 [01:36<01:49, 5.22s/it] |
|
|
49%|████▊ | 19/39 [01:42<01:43, 5.20s/it] |
|
|
51%|█████▏ | 20/39 [01:46<01:33, 4.94s/it] |
|
|
54%|█████▍ | 21/39 [01:50<01:23, 4.64s/it] |
|
|
56%|█████▋ | 22/39 [01:54<01:14, 4.41s/it] |
|
|
59%|█████▉ | 23/39 [01:57<01:03, 3.96s/it] |
|
|
62%|██████▏ | 24/39 [02:00<00:54, 3.66s/it] |
|
|
64%|██████▍ | 25/39 [02:04<00:55, 3.94s/it] |
|
|
67%|██████▋ | 26/39 [02:10<00:56, 4.38s/it] |
|
|
69%|██████▉ | 27/39 [02:14<00:53, 4.47s/it] |
|
|
72%|███████▏ | 28/39 [02:20<00:52, 4.77s/it] |
|
|
74%|███████▍ | 29/39 [02:25<00:50, 5.04s/it] |
|
|
77%|███████▋ | 30/39 [02:31<00:47, 5.25s/it] |
|
|
79%|███████▉ | 31/39 [02:37<00:42, 5.36s/it] |
|
|
82%|████████▏ | 32/39 [02:42<00:38, 5.43s/it] |
|
|
85%|████████▍ | 33/39 [02:47<00:31, 5.24s/it] |
|
|
87%|████████▋ | 34/39 [02:53<00:26, 5.29s/it] |
|
|
90%|████████▉ | 35/39 [02:58<00:21, 5.32s/it] |
|
|
92%|█████████▏| 36/39 [03:03<00:15, 5.23s/it] |
|
|
95%|█████████▍| 37/39 [03:08<00:10, 5.26s/it] |
|
|
97%|█████████▋| 38/39 [03:14<00:05, 5.32s/it] |
|
|
100%|██████████| 39/39 [03:17<00:00, 5.07s/it] |
|
|
Class accuracy is: 81.601883% |
|
|
No obj accuracy is: 97.991463% |
|
|
Obj accuracy is: 75.976616% |
|
|
``` |
|
|
### MAP Calculations |
|
|
```python |
|
|
0%| | 0/39 [00:00<?, ?it/s] |
|
|
3%|▎ | 1/39 [00:40<25:35, 40.40s/it] |
|
|
5%|▌ | 2/39 [01:24<26:05, 42.31s/it] |
|
|
8%|▊ | 3/39 [02:01<24:02, 40.07s/it] |
|
|
10%|█ | 4/39 [02:40<23:04, 39.57s/it] |
|
|
13%|█▎ | 5/39 [03:36<25:45, 45.46s/it] |
|
|
15%|█▌ | 6/39 [04:20<24:45, 45.00s/it] |
|
|
18%|█▊ | 7/39 [05:03<23:37, 44.29s/it] |
|
|
21%|██ | 8/39 [05:47<22:55, 44.36s/it] |
|
|
23%|██▎ | 9/39 [06:33<22:25, 44.84s/it] |
|
|
26%|██▌ | 10/39 [07:06<19:54, 41.20s/it] |
|
|
28%|██▊ | 11/39 [07:58<20:45, 44.49s/it] |
|
|
31%|███ | 12/39 [08:36<19:10, 42.60s/it] |
|
|
33%|███▎ | 13/39 [09:20<18:33, 42.81s/it] |
|
|
36%|███▌ | 14/39 [10:01<17:43, 42.53s/it] |
|
|
38%|███▊ | 15/39 [10:42<16:49, 42.04s/it] |
|
|
41%|████ | 16/39 [11:25<16:10, 42.18s/it] |
|
|
44%|████▎ | 17/39 [12:12<16:02, 43.73s/it] |
|
|
46%|████▌ | 18/39 [12:56<15:20, 43.83s/it] |
|
|
49%|████▊ | 19/39 [13:36<14:12, 42.64s/it] |
|
|
51%|█████▏ | 20/39 [14:20<13:37, 43.04s/it] |
|
|
54%|█████▍ | 21/39 [14:58<12:27, 41.54s/it] |
|
|
56%|█████▋ | 22/39 [15:43<12:01, 42.45s/it] |
|
|
59%|█████▉ | 23/39 [16:29<11:35, 43.49s/it] |
|
|
62%|██████▏ | 24/39 [17:13<10:55, 43.69s/it] |
|
|
64%|██████▍ | 25/39 [18:02<10:34, 45.29s/it] |
|
|
67%|██████▋ | 26/39 [18:41<09:25, 43.53s/it] |
|
|
69%|██████▉ | 27/39 [19:26<08:45, 43.77s/it] |
|
|
72%|███████▏ | 28/39 [20:04<07:44, 42.22s/it] |
|
|
74%|███████▍ | 29/39 [20:45<06:56, 41.65s/it] |
|
|
77%|███████▋ | 30/39 [21:32<06:30, 43.44s/it] |
|
|
79%|███████▉ | 31/39 [22:16<05:47, 43.46s/it] |
|
|
82%|████████▏ | 32/39 [22:52<04:49, 41.32s/it] |
|
|
85%|████████▍ | 33/39 [23:36<04:13, 42.19s/it] |
|
|
87%|████████▋ | 34/39 [24:18<03:29, 41.99s/it] |
|
|
90%|████████▉ | 35/39 [25:00<02:48, 42.17s/it] |
|
|
92%|█████████▏| 36/39 [25:46<02:09, 43.24s/it] |
|
|
95%|█████████▍| 37/39 [26:29<01:26, 43.24s/it] |
|
|
97%|█████████▋| 38/39 [27:18<00:44, 44.74s/it] |
|
|
100%|██████████| 39/39 [27:46<00:00, 42.74s/it] |
|
|
MAP: 0.43667954206466675 |
|
|
``` |
|
|
### Tensorboard Plots |
|
|
**Training Loss vs Steps:**  |
|
|
|
|
|
**Validation Loss vs Steps:** |
|
|
(Info: Validation loss calculated every 10 epochs to save time, thats why the straight line) |
|
|
 |
|
|
|
|
|
### GradCAM Representations |
|
|
EigenCAM is used to generate CAM representation, since usal gradient based method wont work with detection models like Yolo, FRCNN etc. |
|
|
 |
|
|
|
|
|
|
|
|
|
