0
1

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

More than 3 years have passed since last update.

Visual Transformersを動かして、Cifar10で87%は超えた、しかし、その程度。

Posted at

概要

以下のgithubに、Visual Transformersの実装が示されていたので、動かしてみた。いわゆるViTとは、違います。

対象、cifar10とした。

結果

87.31%は出た。↓。

Epoch: 146
[    0/50000 (  0%)]  Loss: 0.1898
[10000/50000 ( 20%)]  Loss: 0.2894
[20000/50000 ( 40%)]  Loss: 0.2176
[30000/50000 ( 60%)]  Loss: 0.2305
[40000/50000 ( 80%)]  Loss: 0.1947
Execution time: 415.95 seconds

Average test loss: 0.4371  Accuracy: 8675/10000 (86.75%)

Epoch: 147
[    0/50000 (  0%)]  Loss: 0.2177
[10000/50000 ( 20%)]  Loss: 0.1389
[20000/50000 ( 40%)]  Loss: 0.1795
[30000/50000 ( 60%)]  Loss: 0.1196
[40000/50000 ( 80%)]  Loss: 0.2700
Execution time: 415.79 seconds

Average test loss: 0.4492  Accuracy: 8617/10000 (86.17%)

Epoch: 148
[    0/50000 (  0%)]  Loss: 0.1744
[10000/50000 ( 20%)]  Loss: 0.2637
[20000/50000 ( 40%)]  Loss: 0.1871
[30000/50000 ( 60%)]  Loss: 0.1839
[40000/50000 ( 80%)]  Loss: 0.1576
Execution time: 415.13 seconds

Average test loss: 0.4565  Accuracy: 8668/10000 (86.68%)

Epoch: 149
[    0/50000 (  0%)]  Loss: 0.2141
[10000/50000 ( 20%)]  Loss: 0.1347
[20000/50000 ( 40%)]  Loss: 0.2369
[30000/50000 ( 60%)]  Loss: 0.1768
[40000/50000 ( 80%)]  Loss: 0.2369
Execution time: 416.12 seconds

Average test loss: 0.4136  Accuracy: 8731/10000 (★87.31%)

まとめ

特にありません。
ViTもそうだが、あまり、いい結果はでない。
コメントなどあれば、お願いします。

リンク

ViT(Vision Transformer)を動かしてみた(4つ)、うっ、Cifar10で80%超えた、ぼちぼちくるかー。すべて他力です。cifar10、imagenet_resized/32x32、結果はまだ「?」付き。

0
1
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
0
1

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?