-
Notifications
You must be signed in to change notification settings - Fork 838
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Error image output with very rare probability #12
Comments
特定の画像で壊れるわけでは無いのですね? |
No, the error occurs randomly, If you do it again on those images, you will get the corrected answers with the same size. So the source is corrected and no error shows up during the calculation. About the conversion, it seems to be occurred on any modes, at least I find scaling and denoising is affected. The my OS is windows 7 64 bit with GTX 960 cuddn, and my friend uses windows 8.1 64 bit with GTX 980ti cuddn. I don't believe the cnn inference calculation is wrong. |
こちらの環境はOS: Windows 8.1 x64, GPU: GTX 660でcuDNNを使っているのですが、出力結果が壊れた画像になったことは一度もないですね… |
I'm still not so sure that this is indeed caused by the program. But I did find this possible bug. You can try processing a large sequence of image that is extracted from a video. And hopefully, this error will reappear 10 times in 100,000 tests. |
GUIで大量の画像を変換すればいいんですね? |
3000ピクセル以上の画像 .WEBP形式に出力することはできません |
@gbaoye ここで議論されている問題とは別の問題なので、新しくIssueを立ててください。 |
出力画像を「D:\1\测试\0000a(RGB)(auto_scale)(Level2)(x2.000000).webp」に書き込めませんでした 2016-03-06 22:50 GMT+08:00 lltcggie notifications@github.com:
|
@gbaoye WebPの幅と高さには最大16383ピクセルに制限されてる、これは仕様です。 ここで議論されている問題とは別の問題なので、新しくIssueを立ててください。 |
Qはどうもありがとうございました 2016-03-07 7:53 GMT+08:00 Heiseikiseki notifications@github.com:
|
こちらで大量の画像変換を試してみたところ、VRAMの使用量が最大ギリギリだとおかしなデータが出力されることがあるようです。 この動作は仕様とするのでIssuesを閉じます。 |
There is a very small probability that you will find that your output image (like .png) might be broken. This happened on my machine and my friend's machine.
The broken image has the corrected file size (this might indicate the calculation process is corrected). The first part of image (basically the top lines of a image) is showing correctly, and the rest of it is broken (like displaying black area or error texture).
The probability is very low, like several errors in 100,000 example. But I found they are more likely to occur within a small period of time, or even continues images.
The text was updated successfully, but these errors were encountered: