Dreampower Versions Save

Fixes

• Fixed issue: checkpoints download path of the zip file was incorrect
• Fixed issue: image size check was not performed when processing a folder as input
• Fixed issue: image modifications were not executed when processing a folder as input
• Fixed issue: images written as output from -a ALTERED feature were not unique, images are now written in their own subfolder, named after the image name and image checksum
• Fixed issue: --auto-resize-crop and --overlay modifications didn't function correctly

Fixes

• Fixed issue: The -a ALTERED feature would write the filenames as 'type.png' instead of the class name of the processing stage
• Fixed issue: If required images did not exist in the directory from -a ALTERED an uncaught exception would be thrown
• Fixed issue: Animated Gifs would fail to process arguments due to bad variable reference

Fixes

• Fixed issue: Check if checkpoint files present error notification and exit prevented the download of checkpoint files via arguments "checkpoints download"

QoL improvements

• More examples in the 'samples' directory, regarding use of directories while using dreampower.

• Transform are now instanciated one time (first use), this gives a huge speed boost when processing multiple images as batch. this feature can be disabled with the new parameter: --disable-persistent-gan

New features

• Parameters are now split into 4 groups: run, checkpoints, gpu-info, daemon

• In daemon mode, dreampower will watch a directory for new images and once detected will automatically process them.

• Input can now have http(s) as source next to locally stored files.

---

Many thanks to @PommeDroid for his contributions.

Use 7-Zip or WinRAR to unzip these files.

QoL improvements

• Auto detect GPU or CPU capability
• Auto detect animated GIF as input
• Improved error handling and messaging
• Checkpoints download via 'dreampower checkpoints download' parameters
• Checkpoints folder location now dynamic via -c CHECKPOINTS parameter
• Version information with -v parameter
• Run-time Debug information with -d parameter

New features

• Color correction with the --color-transfer parameter; transfers the color distribution from the input to the output image using the mean and standard deviations of the Lab color space.

• Bulk / Folder processing; the -i INPUT parameter can now be a folder and any image in that folder and its sub folders will be processed; the -o OUTPUT parameter can now be a folder where the processed images will be saved in, leaving out the -o OUTPUT parameter will cause the output images to be saved in the same folder / sub folder as the input was located; each folder / sub folder can have its own JSON configuration file, default name: 'settings.json'

• Intermediate result images of each of the 6 steps of the algorithm can now be saved via the -a ALTERED parameter that specifies the folder to store them in. The -s STEPS parameter instructs the algorithm to start:end at the specified intermediate steps and use the intermediate result images from the -a ALTERED folder in its execution. This enables manual editing of the intermediate result image files and thus correcting the algorithm, and then have the algorithm continue with the edited / corrected result for the next steps. -s start:end -a ALTERED can be combined with bulk / folder processing.

For developers

• Complete refactoring of the code, making it modular and easier to expand on.

Mirrors

Use 7-Zip to unzip these files.

Improvements

@PommeDroid: Support input resizing & Overlay Json config file for parameters input Simplified setup/build scripts (for developers) @wisp101: Foreign Characters support and Readable Errors GPU support info @1337Roy: Modifiers to overwrite algorithm outcome via input parameters

Use 7-Zip or WinRAR to unzip these files.

Fixes

---

@PommeDroid: Using multi-threading when processing a GIF caused the program to crash.

Use 7-Zip or WinRAR to unzip these files.