Skip to content

Commit 52dc779

Browse files
Coderx7Coderx7
committed
update latest changes in readme
1 parent 3370d81 commit 52dc779

File tree

1 file changed

+65
-62
lines changed

1 file changed

+65
-62
lines changed

‎README.md

Lines changed: 65 additions & 62 deletions
Original file line numberDiff line numberDiff line change
@@ -18,6 +18,9 @@ The pytorch implementation is also very effieicent and the whole model takes onl
1818
#### Update History:
1919

2020
<pre>
21+
-- 2023 Apr 13:
22+
-- new weights for the removed paddings for 1x1 conv layers.
23+
-- some minor fixes
2124
-- 2023 Feb 12:
2225
-- re-structured the repository, moving the old implementation into new directory named 'Cifar` and imagenet into its respective directory
2326
-- updated the old implementation to work with latest version of pytorch.
@@ -33,12 +36,12 @@ The original Caffe implementation can be found here : [Original Caffe implementa
3336

3437
#### ImageNet Result:
3538

36-
| **Method** | **\#Params** | **ImageNet** | **ImageNet-Real-Labels** |
37-
| :--------------------------- | :----------: | :-----------: | :-----------: |
38-
| SimpleNetV1_imagenet(36.33 MB) | 9.5m | 74.17/91.614 | 81.24/94.63 |
39-
| SimpleNetV1_imagenet(21.9 MB) | 5.7m | 71.936/90.3 | 79.12/93.68 |
40-
| SimpleNetV1_imagenet(12.58 MB) | 3m | 68.15/87.762 | 75.66/91.80 |
41-
| SimpleNetV1_imagenet(5.78 MB) | 1.5m | 61.524/83.43 | 69.11/88.10 |
39+
| **Model** | **\#Params** | **ImageNet** | **ImageNet-Real-Labels** |
40+
| :--------------------------- | :----------: | :-----------: | :------------------: |
41+
| simplenetv1_9m_m2(36.3 MB) | 9.5m | 74.23 / 91.748 | 81.22 / 94.756 |
42+
| simplenetv1_5m_m2(22 MB) | 5.7m | 72.03 / 90.324 | 79.328/ 93.714 |
43+
| simplenetv1_small_m2_075(12.6 MB)| 3m | 68.506/ 88.15 | 76.283/ 92.02 |
44+
| simplenetv1_small_m2_05(5.78 MB) | 1.5m | 61.67 / 83.488 | 69.31 / 88.195 |
4245

4346
SimpleNet performs very decently, it outperforms VGGNet, variants of ResNet and MobileNets(1-3)
4447
and its pretty fast as well! and its all using plain old CNN!.
@@ -48,65 +51,65 @@ Small variants of simplenet consistently achieve high performance/accuracy:
4851

4952
| model | samples_per_sec | param_count | top1 | top5 |
5053
|:----------------------------------| :--------------: | :-----------: | :--: | :---: |
51-
|mobilenetv3_small_050 | 3035.37 | 1.59 | 57.89 | 80.194 |
52-
|**simplenetv1_small_m1_05** | 2839.35 | 1.51 | **60.89**|**82.978**|
53-
|lcnet_050 | 2683.57 | 1.88 | 63.1 | 84.382 |
54-
|**simplenetv1_small_m2_05** | 2340.51 | 1.51 |**61.524**|**83.432**|
55-
|mobilenetv3_small_075 | 1781.14 | 2.04 | 65.242 | 85.438 |
56-
|tf_mobilenetv3_small_075 | 1674.31 | 2.04 | 65.714 | 86.134 |
57-
|**simplenetv1_small_m1_075** | 1524.64 | 3.29 |**67.764**|**87.66** |
58-
|tf_mobilenetv3_small_minimal_100 | 1308.27 | 2.04 | 62.908 | 84.234 |
59-
|**simplenetv1_small_m2_075** | 1264.33 | 3.29 |**68.15** |**87.762**|
60-
|mobilenetv3_small_100 | 1263.23 | 2.54 | 67.656 | 87.634 |
61-
|tf_mobilenetv3_small_100 | 1220.08 | 2.54 | 67.924 | 87.664 |
62-
|mnasnet_small | 1085.15 | 2.03 | 66.206 | 86.508 |
63-
|mobilenetv2_050 | 848.38 | 1.97 | 65.942 | 86.082 |
64-
|dla46_c | 531.0 | 1.3 | 64.866 | 86.294 |
65-
|dla46x_c | 318.32 | 1.07 | 65.97 | 86.98 |
66-
|dla60x_c | 298.59 | 1.32 | 67.892 | 88.426 |
54+
|simplenetv1_small_m1_05 | 3100.26 | 1.51 | 61.122 | 82.988 |
55+
|mobilenetv3_small_050 | 3082.85 | 1.59 | 57.89 | 80.194 |
56+
|lcnet_050 | 2713.02 | 1.88 | 63.1 | 84.382 |
57+
|simplenetv1_small_m2_05 | 2536.16 | 1.51 | 61.67 | 83.488 |
58+
|mobilenetv3_small_075 | 1793.42 | 2.04 | 65.242 | 85.438 |
59+
|tf_mobilenetv3_small_075 | 1689.53 | 2.04 | 65.714 | 86.134 |
60+
|simplenetv1_small_m1_075 | 1626.87 | 3.29 | 67.784 | 87.718 |
61+
|tf_mobilenetv3_small_minimal_100 | 1316.91 | 2.04 | 62.908 | 84.234 |
62+
|simplenetv1_small_m2_075 | 1313.6 | 3.29 | 68.506 | 88.15 |
63+
|mobilenetv3_small_100 | 1261.09 | 2.54 | 67.656 | 87.634 |
64+
|tf_mobilenetv3_small_100 | 1213.03 | 2.54 | 67.924 | 87.664 |
65+
|mnasnet_small | 1089.33 | 2.03 | 66.206 | 86.508 |
66+
|mobilenetv2_050 | 857.66 | 1.97 | 65.942 | 86.082 |
67+
|dla46_c | 537.08 | 1.3 | 64.866 | 86.294 |
68+
|dla46x_c | 323.03 | 1.07 | 65.97 | 86.98 |
69+
|dla60x_c | 301.71 | 1.32 | 67.892 | 88.426 |
6770

6871
and this is a sample for larger models: simplenet variants outperform many newer architecures.
6972

70-
| model | samples_per_sec | param_count | top1 | top5 |
71-
|:----------------------------------| :--------------: | :-----------: | :--: | :---: |
72-
| vit_tiny_r_s16_p8_224 | 1882.23 | 6.34 | 71.792 | 90.822 |
73-
| simplenetv1_small_m1_075 | 1516.74 | 3.29 | 67.764 | 87.660 |
74-
| simplenetv1_small_m2_075 | 1260.89 | 3.29 | 68.150 | 87.762 |
75-
| simplenetv1_5m_m1 | 1107.70 | 5.75 | 71.370 | 90.100 |
76-
| deit_tiny_patch16_224 | 991.41 | 5.72 | 72.172 | 91.114 |
77-
| resnet18 | 876.92 | 11.69 | 69.744 | 89.082 |
78-
| simplenetv1_5m_m2 | 835.17 | 5.75 | 71.936 | 90.300 |
79-
| crossvit_9_240 | 602.13 | 8.55 | 73.960 | 91.968 |
80-
| vit_base_patch32_224_sam | 571.37 | 88.22 | 73.694 | 91.010 |
81-
| tinynet_b | 530.15 | 3.73 | 74.976 | 92.184 |
82-
| resnet26 | 524.36 | 16.00 | 75.300 | 92.578 |
83-
| tf_mobilenetv3_large_075 | 505.13 | 3.99 | 73.436 | 91.344 |
84-
| resnet34 | 491.96 | 21.80 | 75.114 | 92.284 |
85-
| regnetx_006 | 478.41 | 6.20 | 73.860 | 91.672 |
86-
| dla34 | 472.49 | 15.74 | 74.620 | 92.072 |
87-
| simplenetv1_9m_m1 | 459.21 | 9.51 | 73.376 | 91.048 |
88-
| repvgg_b0 | 455.36 | 15.82 | 75.160 | 92.418 |
89-
| ghostnet_100 | 407.03 | 5.18 | 73.974 | 91.460 |
90-
| tf_mobilenetv3_large_minimal_100 | 406.84 | 3.92 | 72.250 | 90.630 |
91-
| mobilenetv3_large_100 | 402.08 | 5.48 | 75.766 | 92.544 |
92-
| simplenetv1_9m_m2 | 389.94 | 9.51 | 74.170 | 91.614 |
93-
| tf_mobilenetv3_large_100 | 388.30 | 5.48 | 75.518 | 92.604 |
94-
| mobilenetv2_100 | 295.68 | 3.50 | 72.970 | 91.020 |
95-
| densenet121 | 293.94 | 7.98 | 75.584 | 92.652 |
96-
| mnasnet_100 | 262.25 | 4.38 | 74.658 | 92.112 |
97-
| vgg11 | 260.38 | 132.86 | 69.028 | 88.626 |
98-
| vgg11_bn | 248.92 | 132.87 | 70.360 | 89.802 |
99-
| mobilenetv2_110d | 230.80 | 4.52 | 75.038 | 92.184 |
100-
| efficientnet_lite0 | 224.81 | 4.65 | 75.476 | 92.512 |
101-
| tf_efficientnet_lite0 | 219.93 | 4.65 | 74.832 | 92.174 |
102-
| vgg13 | 154.03 | 133.05 | 69.926 | 89.246 |
103-
| vgg13_bn | 144.39 | 133.05 | 71.594 | 90.376 |
104-
| vgg16 | 123.70 | 138.36 | 71.590 | 90.382 |
105-
| vgg16_bn | 117.06 | 138.37 | 73.350 | 91.504 |
106-
| vgg19 | 103.71 | 143.67 | 72.366 | 90.870 |
107-
| vgg19_bn | 98.59 | 143.68 | 74.214 | 91.848 |
108-
109-
Benchmark was done using a GTX1080 on Pytorch 1.11 with fp32, nhwc, batchsize of 256, input size = `224x224x3`.
73+
| model | samples_per_sec | param_count | top1 | top5 |
74+
|:----------------------------------| :--------------: | :-----------: | :----: | :----: |
75+
| simplenetv1_small_m1_075 | 3036.82 | 3.29 | 67.784 | 87.718 |
76+
| simplenetv1_small_m2_075 | 2589.4 | 3.29 | 68.506 | 88.15 |
77+
| vit_tiny_r_s16_p8_224 | 2419.27 | 6.34 | 71.792 | 90.822 |
78+
| simplenetv1_5m_m1 | 2204.86 | 5.75 | 71.548 | 89.94 |
79+
| resnet18 | 1824.39 | 11.69 | 69.744 | 89.082 |
80+
| simplenetv1_5m_m2 | 1821.86 | 5.75 | 72.03 | 90.324 |
81+
| regnetx_006 | 1710.9 | 6.2 | 73.86 | 91.672 |
82+
| mobilenetv3_large_100 | 1568.72 | 5.48 | 75.766 | 92.544 |
83+
| tf_mobilenetv3_large_minimal_100 | 1564.4 | 3.92 | 72.25 | 90.63 |
84+
| tf_mobilenetv3_large_075 | 1560.84 | 3.99 | 73.436 | 91.344 |
85+
| ghostnet_100 | 1447.97 | 5.18 | 73.974 | 91.46 |
86+
| tinynet_b | 1413.59 | 3.73 | 74.976 | 92.184 |
87+
| tf_mobilenetv3_large_100 | 1396.66 | 5.48 | 75.518 | 92.604 |
88+
| mnasnet_100 | 1238.11 | 4.38 | 74.658 | 92.112 |
89+
| mobilenetv2_100 | 1158.3 | 3.5 | 72.97 | 91.02 |
90+
| simplenetv1_9m_m1 | 1089.47 | 9.51 | 73.792 | 91.486 |
91+
| resnet34 | 1071.63 | 21.8 | 75.114 | 92.284 |
92+
| efficientnet_lite0 | 1020.15 | 4.65 | 75.476 | 92.512 |
93+
| deit_tiny_patch16_224 | 1007.63 | 5.72 | 72.172 | 91.114 |
94+
| simplenetv1_9m_m2 | 933.93 | 9.51 | 74.23 | 91.748 |
95+
| tf_efficientnet_lite0 | 920.71 | 4.65 | 74.832 | 92.174 |
96+
| dla34 | 871.88 | 15.74 | 74.62 | 92.072 |
97+
| mobilenetv2_110d | 867.18 | 4.52 | 75.038 | 92.184 |
98+
| resnet26 | 809.01 | 16 | 75.3 | 92.578 |
99+
| repvgg_b0 | 788.45 | 15.82 | 75.16 | 92.418 |
100+
| crossvit_9_240 | 618.98 | 8.55 | 73.96 | 91.968 |
101+
| vgg11 | 602.05 | 132.86 | 69.028 | 88.626 |
102+
| vgg11_bn | 522.94 | 132.87 | 70.36 | 89.802 |
103+
| vit_base_patch32_224_sam | 514.09 | 88.22 | 73.694 | 91.01 |
104+
| densenet121 | 460.65 | 7.98 | 75.584 | 92.652 |
105+
| vgg13 | 373.71 | 133.05 | 69.926 | 89.246 |
106+
| vgg13_bn | 326.26 | 133.05 | 71.594 | 90.376 |
107+
| vgg16 | 312.99 | 138.36 | 71.59 | 90.382 |
108+
| vgg16_bn | 276.35 | 138.37 | 73.35 | 91.504 |
109+
| vgg19 | 269.28 | 143.67 | 72.366 | 90.87 |
110+
| vgg19_bn | 239.6 | 143.68 | 74.214 | 91.848 |
111+
112+
Benchmark was done using a GTX1080 on Pytorch 1.11 with fp32, nchw, batchsize of 256, input size = `224x224x3`.
110113
For all benchmark results [look here](https://github.com/Coderx7/SimpleNet_Pytorch/tree/master/ImageNet/training_scripts/imagenet_training/results)
111114

112115
-- The models pretrained weights (pytorch, onnx, jit) can be found in [Release section](https://github.com/Coderx7/SimpleNet_Pytorch/releases)

0 commit comments

Comments
 (0)