Skip to content

Commit 918479f

Browse files
Coderx7Coderx7
authored
Update README.md
1 parent d855595 commit 918479f

File tree

1 file changed

+128
-15
lines changed

1 file changed

+128
-15
lines changed

‎README.md‎

Lines changed: 128 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -188,21 +188,134 @@ CIFAR100 Extended results:
188188
** Achieved using several data-augmentation tricks
189189

190190
**Flops and Parameter Comparison:**
191-
192-
| | MACC | COMP | ADD | DIV | EXP | Activations | Params | SIZE(MB) |
193-
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
194-
| SimpleNet | 652 M | 0.838M | 10 | 10 | 10 | 1 M | 5 M | 20.9 |
195-
| SqueezeNet | 861 M | 10 M | 226K | 1.51M | 1K | 13 M | 1 M | 4.7 |
196-
| Inception v4\* | 12270 M | 21.9 M | 5.34M | 897K | 1K | 73 M | 43 M | 163 |
197-
| Inception v3\* | 5710 M | 16.5 M | 2.59M | 1.71M | 11K | 33 M | 24 M | 91 |
198-
| Inception-ResNetv2\* | 9210 M | 17.6 M | 2.36M | 1K | 1K | 74 M | 32 M | 210 |
199-
| ResNet-152 | 11300 M | 22.33M | 35.27M | 22.03M | 1K | 100.26M | 60.19M | 230 |
200-
| ResNet-50 | 3870 M | 10.9 M | 1.62M | 1.06M | 1K | 47 M | 26 M | 97.70 |
201-
| AlexNet | 1140 M | 1.77 M | 478K | 955K | 478K | 2 M | 62 M | 217.00 |
202-
| GoogleNet | 1600 M | 16.1 M | 883K | 166K | 833K | 10 M | 7 M | 22.82 |
203-
| Network in Network | 1100 M | 2.86 M | 370K | 1K | 1K | 3.8 M | 8 M | 29 |
204-
| VGG16 | 15740 M | 19.7 M | 1K | 1K | 1K | 29 M | 138 M | 512.2 |
205-
191+
<table>
192+
<caption>Flops and Parameter Comparison of Models trained on ImageNet</caption>
193+
<thead>
194+
<tr class="header">
195+
<th style="text-align: left;"><strong>Model</strong></th>
196+
<th style="text-align: center;"><strong>MACC</strong></th>
197+
<th style="text-align: center;"><strong>COMP</strong></th>
198+
<th style="text-align: center;"><strong>ADD</strong></th>
199+
<th style="text-align: center;"><strong>DIV</strong></th>
200+
<th style="text-align: center;"><strong>Activations</strong></th>
201+
<th style="text-align: center;"><strong>Params</strong></th>
202+
<th style="text-align: center;"><strong>SIZE(MB)</strong></th>
203+
</tr>
204+
</thead>
205+
<tbody>
206+
<tr class="odd">
207+
<td style="text-align: left;">SimpleNet</td>
208+
<td style="text-align: center;">1.9G</td>
209+
<td style="text-align: center;">1.82M</td>
210+
<td style="text-align: center;">1.5M</td>
211+
<td style="text-align: center;">1.5M</td>
212+
<td style="text-align: center;">6.38M</td>
213+
<td style="text-align: center;">6.4M</td>
214+
<td style="text-align: center;">24.4</td>
215+
</tr>
216+
<tr class="even">
217+
<td style="text-align: left;">SqueezeNet</td>
218+
<td style="text-align: center;">861.34M</td>
219+
<td style="text-align: center;">9.67M</td>
220+
<td style="text-align: center;">226K</td>
221+
<td style="text-align: center;">1.51M</td>
222+
<td style="text-align: center;">12.58M</td>
223+
<td style="text-align: center;">1.25M</td>
224+
<td style="text-align: center;">4.7</td>
225+
</tr>
226+
<tr class="odd">
227+
<td style="text-align: left;">Inception v4*</td>
228+
<td style="text-align: center;">12.27G</td>
229+
<td style="text-align: center;">21.87M</td>
230+
<td style="text-align: center;">53.42M</td>
231+
<td style="text-align: center;">15.09M</td>
232+
<td style="text-align: center;">72.56M</td>
233+
<td style="text-align: center;">42.71M</td>
234+
<td style="text-align: center;">163</td>
235+
</tr>
236+
<tr class="even">
237+
<td style="text-align: left;">Inception v3*</td>
238+
<td style="text-align: center;">5.72G</td>
239+
<td style="text-align: center;">16.53M</td>
240+
<td style="text-align: center;">25.94M</td>
241+
<td style="text-align: center;">8.97M</td>
242+
<td style="text-align: center;">41.33M</td>
243+
<td style="text-align: center;">23.83M</td>
244+
<td style="text-align: center;">91</td>
245+
</tr>
246+
<tr class="odd">
247+
<td style="text-align: left;">Incep-Resv2*</td>
248+
<td style="text-align: center;">13.18G</td>
249+
<td style="text-align: center;">31.57M</td>
250+
<td style="text-align: center;">38.81M</td>
251+
<td style="text-align: center;">25.06M</td>
252+
<td style="text-align: center;">117.8M</td>
253+
<td style="text-align: center;">55.97M</td>
254+
<td style="text-align: center;">214</td>
255+
</tr>
256+
<tr class="even">
257+
<td style="text-align: left;">ResNet-152</td>
258+
<td style="text-align: center;">11.3G</td>
259+
<td style="text-align: center;">22.33M</td>
260+
<td style="text-align: center;">35.27M</td>
261+
<td style="text-align: center;">22.03M</td>
262+
<td style="text-align: center;">100.11M</td>
263+
<td style="text-align: center;">60.19M</td>
264+
<td style="text-align: center;">230</td>
265+
</tr>
266+
<tr class="odd">
267+
<td style="text-align: left;">ResNet-50</td>
268+
<td style="text-align: center;">3.87G</td>
269+
<td style="text-align: center;">10.89M</td>
270+
<td style="text-align: center;">16.21M</td>
271+
<td style="text-align: center;">10.59M</td>
272+
<td style="text-align: center;">46.72M</td>
273+
<td style="text-align: center;">25.56M</td>
274+
<td style="text-align: center;">97.70</td>
275+
</tr>
276+
<tr class="even">
277+
<td style="text-align: left;">AlexNet</td>
278+
<td style="text-align: center;">7.27G</td>
279+
<td style="text-align: center;">17.69M</td>
280+
<td style="text-align: center;">4.78M</td>
281+
<td style="text-align: center;">9.55M</td>
282+
<td style="text-align: center;">20.81M</td>
283+
<td style="text-align: center;">60.97M</td>
284+
<td style="text-align: center;">217.00</td>
285+
</tr>
286+
<tr class="odd">
287+
<td style="text-align: left;">GoogleNet</td>
288+
<td style="text-align: center;">16.04G</td>
289+
<td style="text-align: center;">161.07M</td>
290+
<td style="text-align: center;">8.83M</td>
291+
<td style="text-align: center;">16.64M</td>
292+
<td style="text-align: center;">102.19M</td>
293+
<td style="text-align: center;">7M</td>
294+
<td style="text-align: center;">40</td>
295+
</tr>
296+
<tr class="even">
297+
<td style="text-align: left;">NIN</td>
298+
<td style="text-align: center;">11.06G</td>
299+
<td style="text-align: center;">28.93M</td>
300+
<td style="text-align: center;">380K</td>
301+
<td style="text-align: center;">20K</td>
302+
<td style="text-align: center;">38.79M</td>
303+
<td style="text-align: center;">7.6M</td>
304+
<td style="text-align: center;">29</td>
305+
</tr>
306+
<tr class="odd">
307+
<td style="text-align: left;">VGG16</td>
308+
<td style="text-align: center;">154.7G</td>
309+
<td style="text-align: center;">196.85M</td>
310+
<td style="text-align: center;">10K</td>
311+
<td style="text-align: center;">10K</td>
312+
<td style="text-align: center;">288.03M</td>
313+
<td style="text-align: center;">138.36M</td>
314+
<td style="text-align: center;">512.2</td>
315+
</tr>
316+
</tbody>
317+
</table>
318+
206319
\*Inception v3, v4 and Inception-ResNetV2 did not have any Caffe model, so we reported their size related information from mxnet2 and tensorflow3 respectively. Inception-ResNet-V2 would take 60 days of training with 2 Titan X to achieve the reported accuracy4
207320

208321
As it can be seen, our architecture both has much fewer number of parameters (with an exception of squeezenet) and also much fewer number of operations.

0 commit comments

Comments
 (0)