Releases: Coderx7/SimpleNet_Pytorch
Release list
ImageNet pretrained weights
Here are the latest pretrained weights on ImageNet dataset.
In this version, we fixed an unintentional bug in the model, where the 1x1 convs had paddings of 1,
this is now taken care of and new weightsand models are provided.
m2 variants:
| Method | #Params | ImageNet | ImageNet-Real-Labels |
|---|---|---|---|
| simplenetv1_9m_m2(36.3 MB) | 9.5m | 74.23/91.748 | 81.22/94.756 |
| simplenetv1_5m_m2(22 MB) | 5.7m | 72.03/90.324 | 79.328/93.714 |
| simplenetv1_small_m2_075(12.6 MB) | 3m | 68.506/88.15 | 76.283/92.02 |
| simplenetv1_small_m2_05(5.78 MB) | 1.5m | 61.67/83.488 | 69.31/ 88.195 |
m1 variants:
| Method | #Params | ImageNet | ImageNet-Real-Labels |
|---|---|---|---|
| simplenetv1_9m_m1(36.3 MB) | 9.5m | 73.792/91.486 | 81.196/94.512 |
| simplenetv1_5m_m1(22 MB) | 5.7m | 71.548/89.94 | 79.076/93.36 |
| simplenetv1_small_m1_075(12.6 MB) | 3m | 67.784/87.718 | 75.448/91.69 |
| simplenetv1_small_m1_05(5.78 MB) | 1.5m | 61.122/82.988 | 68.58/87.64 |
Note1:
some of these weights are achieved from finetuning the previous weights to shorten the training time
some others are trained from scratch.
imagenet pretrained weights with nopad in conv1x1 layers
These are the weights finetuned on previously trained weights, with the exception that the last 2 conv1x1 layers now have no padding
this allows for faster inference and I tried to improve the accuracy and here are the models so far.
the sample quantized weights are statically quantized, in order to get full accuracy like the one in full precision weights, they must be fintuned in QAT.
Initial ImageNet Models
Initial ImageNet pretrained weights are now available for download.
m2 variants:
| Method | #Params | ImageNet | ImageNet-Real-Labels |
|---|---|---|---|
| simplenetv1_9m_m2(36.33 MB) | 9.5m | 74.17/91.61 | 81.24/94.63 |
| simplenetv1_5m_m2(21.9 MB) | 5.7m | 71.94/90.3 | 79.12/93.68 |
| simplenetv1_small_m2_075(12.58 MB) | 3m | 68.15/87.76 | 75.66/91.80 |
| simplenetv1_small_m2_05(5.78 MB) | 1.5m | 61.53/83.43 | 69.11/88.10 |
m1 variants:
| Method | #Params | ImageNet | ImageNet-Real-Labels |
|---|---|---|---|
| simplenetv1_9m_m1(36.33 MB) | 9.5m | 73.376/91.048 | 80.72/94.26 |
| simplenetv1_5m_m1(21.9 MB) | 5.7m | 71.37/90.10 | 78.77/93.50 |
| simplenetv1_small_m1_075(12.58 MB) | 3m | 67.764/87.66 | 75.48/91.66 |
| simplenetv1_small_m1_05(5.78 MB) | 1.5m | 60.89/82.978 | 68.46/87.64 |
Initial ImageNet models
Initial ImageNet models that I trained recently