Large Images Shrink (and crashes Deep Dreamer)



  • Hi,

    I have an image that's 4096x2048 pixels, but when I try rendering with Deep Dreamer it most often crashes, but when it does render the resulting image is smaller and the size varies randomly it seems. Is there a limitation on the image size, or a setting somewhere?

    This is from the Console:
    12/18/15 9:33:43.977 AM Deep Dreamer[13684]: [CL_INVALID_VALUE] : OpenCL Error : Failed to retrieve device information! Insufficient memory to hold value! Need 24 bytes.
    12/18/15 9:34:08.773 AM Deep Dreamer[13684]: Handling crash with signal 6...
    12/18/15 9:34:21.636 AM Deep Dreamer[13684]: BUG in libdispatch client: kevent[EVFILT_MEMORYSTATUS] add: "Operation not supported" - 0x2d
    12/18/15 9:34:22.233 AM Deep Dreamer[13684]: assertion failed: 15B42: libxpc.dylib + 79134 [3E09C275-A33B-357A-B0AB-A2DDF88EC9D5]: 0x89


  • administrators

    Hey!

    I have seen Deep Dreamer handling images of similar size or larger than 4096x2048, however I wouldn't recommend it. The processing time and memory requirements are quite immense for larger images.

    The resulting image size being smaller than the source is quite normal. Deep Dreamer samples the image at different sizes according to the Octaves and Octave Scale parameters.

    If you do get a crash, try lowering the size a little to see if that helps.

    Cheers
    Tom



  • Thanks for the reply. My interest in using Deep Dreamer, and why I purchased it, is to process high resolution images. That it's acting unpredictably, creating different sized images and crashing, is making it very difficult to use. What I expect is for the app to fail gracefully with an error message when the app runs out of memory, not a crash where I then lose my work in progress. I also expect that the output image dimensions will be the same as the image I give it.

    BTW, I've tried running it with "CPU Normal" and "CPU Fast", in case it's a GPU issue, but both result in the error message:
    Corruption Detected!
    These nightmares have breached my internals and started causing blackouts! Please try another mode.



  • Can you let me know the max size we should expect the software to handle? The sales page didn't spec and max size output. Is there any chance this will be fixed?


  • administrators

    Hi ya, that is actually proving a difficult question to answer. The size of image that works depends on a number of factors including the GPU being used and the memory available. As mentioned above we have regularly processed images on that are 4096x2048 but this can consume a huge amount of memory due to the way the sampling is done in the Google Deep Dream engine.

    We have found that our users get much more consistent results using smaller images of 2048x2048 and smaller.

    We will add this information to the website as it is an important piece of info.

    I hope that helps.

    Mike



  • @Mike said:

    but this can consume a huge amount of memory due to the way the sampling is done in the Google Deep Dream engine.

    Thanks Mike!

    I know you aren't responsible for the Google's Deep Dream code... but is there any chance your time might improve upon it and make it work better for high res images? Please consider this a feature request. Thanks so much for this app, I'm having a lot of fun with it!


  • administrators

    Thanks for the feedback. It's great to hear you are enjoying the app. That's the best feedback we can get :)

    We'll continue to monitor changes to the Deep Dream engine and work on enhancements to Deep Dreamer.

    Regards

    Mike



  • I am in a similar position. I love the possibilities, but the output is horrendously small and useless for print. I got this app as part of a bundle, so no big sweat, but there is no way to get decent resolution out of it, and no controls for that at all (sizing output, image type, etc.). This is NOT a pro level app, nor ready for prime time yet. Sad that.


Log in to reply
 

Looks like your connection to Forum was lost, please wait while we try to reconnect.