When using overlay scaling the maskdet will be exported instead of the maskfin if you've got nf & transform selected as transform method. If I use scale crop/pad it works correctly, but overlay is better in most scenarios.

V1.3
latest checkpoints & dreampower

The output image colors are darker and more dull than the input image for all pixels.

Correct this by sampling the input image and the output image and reverse the change for all pixels.

Latest versions of MacOS require command-line-tools and their libraries to be signed by the developer.

A work around is for the user to white list the tool, but this is not obvious to the average user.

https://github.com/pyinstaller/pyinstaller/wiki/Recipe-OSX-Code-Signing
Shows how the code signing would work, this would need to be implemented in the CI after the package was created by PyInstaller.

https://imgur.com/3D3QaUl

Just like that - no error. anything up to date

Describe the bug
What's with the colors output? The images come out with different hue, the saturation is overdone and so are the white and black values (both clipping). The results is even worse than that, the blacks are actually grey, there is also some colored noise/artifacts/dithering in and around those darker areas (pubic area,shadows, etc). There are couple issues that may cause it but there has to be an compression/low bit depth/color depth one... Also why is the algorithm even affecting the rest of the image? It generates masks so why would it even change the colors of the rest?

How to reproduce: just make any fake using it, happens 100%, more or less severely.

Version: 1.1.0

Also will there be some kind of seamless method of just putting the faked areas over the original image (using those masks it generates) and support for higher resolutions (as it is the app isn't demanding so I could see it handling 1024x1024 resolution easily).

Looking forward to a fix for it.

Copying this over from the Dreamtime Issue Tracker. @deeppppp suggested in this issue that we support python 3.7 rather than 3.6, as pytorch makes wheels for that version too. This seems better imo than depending on any python version > 3.6.2. However, this would also make it harder to build on some Linux OSes where python 3.7 isn't in the package manager.

Verify input parameter values to catch exceptions before they occur.
for example:
-o OUTPUT -> OUTPUT needs to be a file name with an extension of an image format (png / jpg)

verify directory values are indeed directories
filename values are existing files
image filenames are images
JSON filenames are JSON files
etc.

The current archives for v1.0.0 are missing the input.png sample image.
causing running dreampower without any parameters to exit with:

Error : input.png file doesn't exist
Traceback (most recent call last):
File "D:\DeepNude\DreamNet\DreamPower\dreampower\main.py", line 346, in
File "D:\DeepNude\DreamNet\DreamPower\dreampower\main.py", line 192, in main
NameError: name 'exit' is not defined
[6836] Failed to execute script main

solution: input.png will be added to the archives and those will be updated in Mega / MediaFire / Github

I would like to know how to run in some way in google colab.
I wait your answer.

So, I'm guessing we all got banned?

If the input image is not 512x512 and none of the resizing options are selected, an user friendly error should be shown, currently it crashes on the first GAN step.

Whenever I try using the app, the picture is processing smoothly, but afterward I get hit with this error:
"DreamPower has reported that the photo has been transformed but the file does not exist! This may be due to a problem saving the photo, verify that DreamPower has write permissions and that your Antivirus is not detecting false positives. It is also possible that DreamPower is ending abruptly due to a major problem."

What can I do to fix it? It does have write permissions, antivirus isn't detecting anything...

The attached image generates green boobs.
How to prevent this?
Does the input file need color adjustment?
Can't imagine better pose than this.

Using latest release, CPU only mode.

It would be nice having a slider where you can adjust the amount of muscle. Hopefully you'll consider this.

You are doing god's work.

the other flags like -h, -i, -o, --cpu, --cpu, etc. but there are none of the other options under the "Differences with DeepNude" header on the readme

Guide for training the model would be nice. Would geatly improve quality of the pictures.

Says update available: 0.0.1 but update button doesn't work.

Steps to reproduce the behavior:

1. Go to checkpoints
2. Click on update
3. Scroll down to Dreampower update available
4. Click on update
5. See Update failed
   There was a problem trying to download the update, please verify that you are connected to the Internet. If the problem persists try to configure or temporarily disable your Firewall/Antivirus/VPN.

I don't have any firewalls or antivirus up and am obviously connected to the internet.

Can't use without updating.

I have added the folder c:\dreampower to my Windows PATH environment variable so that I can type dreampower in the CLI from anywhere. However it throws the error that the checkpoints can not be found and need to be downloaded. I believe this is a bug.

To Reproduce
Steps to reproduce the behavior:

1. Download DreamPower prebuilt v1.2.3 for Windows (GPU)
2. Extract it and add the folder to your Windows PATH environment variable
3. Go to some other folder like desktop and type dreampower
4. See error

Expected behavior
It should find the checkpoints relative to itself and not the current process directory. (cwd) However it should not look for the input image itself in its own directory. The whole point is that I can run this in another folder.

Enviroment

• Version [1.2.3]
• OS: [e.g. Windows 10 1803]

Additional context
v1.0.0 works fine.

Hello,

When I try to use:
dreampower checkpoints download
I get that error:

Downloading https://cdn.dreamnet.tech/releases/checkpoints/v0.0.1.zip
[INFO] Extracting C:\Users\user\AppData\Local\Temp\tmpbf5e3x1e\v0.0.1.zip
[ERROR] stat: path should be string, bytes, os.PathLike or integer, not dict
[ERROR] Something Gone Bad Download Downloading The Checkpoints

Enviroment:

• Version 1.2.2
• OS: Windows 10 1909

Thanks.

hallo,I wanted to make a suggestion. Since version 3.5 or something you have made it mandatory that there is a fake areola and nipple on all results. The minimum is 0,3 that we can choose.before that with a preference of 0.0 there were a lot of cases when the model was close enough where we could achieve a transparency effect with 0.0 you could make out exactly where and how big the areola was. if you tinker a little with the result with a third party program like gimp you could achieve excellent results.now with the fake nipple it is unesthaitic and confusing.I think the program has lost since the mandatory insertion of 0.3 and I don't understand the reason for this rule anyway.

Please restore the option of 0.0 on nipple and areola thank you very much for your excellent work!!

Describe the bug
If you try to process an image with an extension written in capital letters (e.g. image.JPG) dreampower says the file format is not supported. If the name is changed to image.jpg it works fine again.

To Reproduce
Steps to reproduce the behavior:

1. Take image name.JPG
2. Process
3. Get error

Expected behavior
Checking of extension should be case-insensitive

Enviroment (please complete the following information):

• DreamPower Version 1.2.3
• OS: Win10

DreamPower is a fork of deepnude_official but with constant improvements from the developers of DreamNet and the world.

How constant exactly are these improvements? Since I forked this repo, the checkpoints remain at version 0.0.1, I assume new improvements will be pushed to 0.0.2 and do not overwrite 0.0.1?

Maybe it is not a bad idea to build in a small checker into the CLI tool that will tell you when its time to update.

When I try to run main.py in virtual environment get error as below. I am not sure what script is suggesting by ' download them using main.py'

(env2) PS F:\dream> python main.py --input Img.jpg -a ALTERED -s 0:1
usage: main.py [-h] [-d] [-i INPUT] [-o OUTPUT] [--cpu | --gpu GPU]
[--bsize BSIZE] [--asize ASIZE] [--nsize NSIZE] [--vsize VSIZE]
[--hsize HSIZE] [-n N_RUNS] [--n-cores N_CORES]
[--auto-resize | --auto-resize-crop | --auto-rescale | --overlay OVERLAY | --ignore-size]
[--color-transfer] [-s STEPS] [-a ALTERED] [-c CHECKPOINTS]
[-j JSON_ARGS] [--json-folder-name JSON_FOLDER_NAME] [-v]
{gpu-info,checkpoints} ...
main.py: error: Checkpoints file not found in directory F:\dream\checkpoints. You can download them using : main.py checkpoints download

Anyway we could reduce the required memory from 8GB to 4GB with a parameter?
Even if it costs more time to run.
It would make the SW more accessible to people (not talking about 1st world people).

There is a malloc somewhere for 3GB, but most of the time it only uses 1-2GB RAM.

I assume we load the trained model into RAM (3x700MB), what if we will grow by 10x?
One 512x512 frame only costs 256KB, which fits 12200 times into 3GB, is that mandatory?

(without CUDA use case)

Transformation #1 failed
The process has been interrupted by an unknown error, this may be caused by a corrupt installation, please check the console for more information.

For more information please take a screenshot and report the following on Github or here:

    Error: usage: dreampower [-h] [-d] [-v] {run,checkpoints,gpu-info,daemon} ...

dreampower: error: argument mode: invalid choice: 'D:\Users\psych\AppData\Local\Programs\DreamTime\cli/624195e5-84cc-44f0-86a2-0b2cafb22a40.png' (choose from 'run', 'checkpoints', 'gpu-info', 'daemon')

usage: dreampower [-h] [-d] [-v] {run,checkpoints,gpu-info,daemon} ... dreampower: error: argument mode: invalid choice: 'D:\Users\psych\AppData\Local\Programs\DreamTime\cli/624195e5-84cc-44f0-86a2-0b2cafb22a40.png' (choose from 'run', 'checkpoints', 'gpu-info', 'daemon')

Do you planning support for AMD cards using OpenCL platform?

The CLI doesn't appear to check if the image saved correctly and returns 0 even if it failed to save.

To gain more supporters of the app and not being tagged as machism, this app should support male images as well as female. It could be customized with dick size and pubic hair. This maybe could allow in the future to make TS transformations. Male dressed to naked female and female dressed to naked male.

Would it be possible to add a stepwise function such that manual changes to the masks (presumably maskref and maskfin) could be run through the program?

Would like to see if this would allow for easy re-positioning of nipples, a common mistake of the algorithm etc

Allow -i INPUT to be a folder and traverse this folder down including any sub folders and nudify any image file encountered.

if a JSON file is in the folder, try to process that as if it was given with the -j JSON_ARGS parameter for each file in the folder (overwrites a given -j JSON_ARGS parameter) and its sub folders until a new JSON file is encountered down the folder tree.

if -o OUTPUT is given and this is a folder, then place the nudify output images in that folder, recreating any sub folders from the input folder if sub folders where encountered in the input folder.

if -o OUTPUT is not a folder or is missing, then write the output images to the input folder, same naming scheme as the encountered input file with a postfix '_out' or similar.

if -s STEPS and -a ALTERED are given place the intermediate files in the ALTERED folder, start from step 0, end with the ending step and write this ending step as the output image

Traceback (most recent call last):
File "main.py", line 82, in
argv.run()
File "/content/dreampower/argv/init.py", line 35, in run
args.func(args)
File "/content/dreampower/main.py", line 25, in main
select_processing().run()
File "/content/dreampower/processing/init.py", line 23, in run
r = self._execute(args)
File "/content/dreampower/processing/image.py", line 60, in _execute
r = run_worker(p, self.__image_steps, config=self._args)
File "/content/dreampower/processing/worker.py", line 49, in run_worker
r = w.run([image_step[i] for i in w.input_index], config=config)
File "/content/dreampower/transform/init.py", line 26, in run
return super().run(*args, config=config)
File "/content/dreampower/processing/init.py", line 23, in run
r = self._execute(*args)
File "/content/dreampower/transform/opencv/mask.py", line 107, in _execute
bodypart_list = extract_annotations(args[1], enable_pubes)
File "/content/dreampower/transform/opencv/bodypart/extract.py", line 19, in extract_annotations
tits_list = find_body_part(maskdet, "tit")
File "/content/dreampower/transform/opencv/bodypart/extract.py", line 64, in find_body_part
contours, _ = cv2.findContours(color_mask, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
ValueError: too many values to unpack (expected 2)

Ubuntu 16, all work but no ignore size parameter

[INFO] Executing Mask To Maskref
Traceback (most recent call last):
File "DeepNude/main.py", line 82, in
argv.run()
File "/var/www/html/.../.../argv/init.py", line 35, in run
args.func(args)
File "/var/www/html/.../.../main.py", line 25, in main
select_processing().run()
File "/var/www/html/.../.../processing/init.py", line 23, in run
r = self._execute(args)
File "/var/www/html/.../.../processing/image.py", line 96, in _execute
r = run_worker(p, self.__image_steps, config=self._args)
File "/var/www/html/.../.../processing/worker.py", line 49, in run_worker
r = w.run([image_step[i] for i in w.input_index], config=config)
File "/var/www/html/.../.../transform/init.py", line 26, in run
return super().run(*args, config=config)
File "/var/www/html/.../.../processing/init.py", line 23, in run
r = self._execute(*args)
File "/var/www/html/.../.../transform/opencv/mask.py", line 55, in _execute
res2 = cv2.bitwise_and(green, green, mask=green_mask)
cv2.error: OpenCV(4.2.0) /io/opencv/modules/core/src/arithm.cpp:250: error: (-215:Assertion failed) (mtype == CV_8U || mtype == CV_8S) && _mask.sameSize(*psrc1) in function 'binary_op'

Describe the bug
The second or third time I run this application prebuilt version, I receive an error that CUDA is out of memory.

To Reproduce
Steps to reproduce the behavior:

1. Download and extract binaries for Windows
2. Install the endpoints using dreampower.exe
3. Run dreampower.exe a few times on different images
4. See error

Expected behavior
I expect the cache of my GPU to be cleared after the process is finished and/or before it starts.

Enviroment

• Version: latest
• OS: Windows 10 x64
• GPU: NVIDIA GeForce GTX 1050

Additional context

RuntimeError: CUDA out of memory. Tried to allocate 36.00 MiB (GPU 0; 2.00 GiB total capacity; 564.01 MiB already allocated; 17.85 MiB free; 15.99 MiB cached)

https://github.com/ROCmSoftwarePlatform/MIOpen

The process has been interrupted by an unknown error, this may be caused by a corrupt installation, please check the console for more information.

For more information please take a screenshot and report the following on Github or here:

    Error: [INFO] Welcome to DreamPower

[INFO] GAN Processing Will Use GPU IDs: [0]

[INFO] Executing Image Processing [INFO] Processing on C:\Users\Eric Tsai.LAPTOP-PEHSU7T7\Desktop/1566161344440605.jpg

[INFO] Executing Image To Crop [INFO] Executing Image To Resized

Traceback (most recent call last): File "main.py", line 82, in File "argv_init__.py", line 35, in run File "main.py", line 25, in main File "processing_init.py", line 23, in run File "processing\image.py", line 78, in execute File "processing\worker.py", line 49, in runworker File "transform__init.py", line 26, in run File "processing_init__.py", line 23, in run File "transform\opencv\resize.py", line 76, in execute

File "transform\opencv\resize.py", line 83, in calculatenew_size ZeroDivisionError: float division by zero [4960] Failed to execute script main

Describe the bug
[6983] Error loading Python lib '/media/pixiepox/ssd_storage/DreamPower/libpython3.7m.so.1.0': dlopen: /lib/x86_64-linux-gnu/libc.so.6: version `GLIBC_2.25' not found (required by /media/pixiepox/ssd_storage/DreamPower/libpython3.7m.so.1.0)

To Reproduce
Steps to reproduce the behavior:

1. I am currently on DeepinOS 15.11
2. Whatever command I try to run is met with this error.
3. PS - I tried the software on Ubuntu and Mint previously and it worked 100% fine.

Screenshots
If applicable, add screenshots to help explain your problem.

Enviroment (please complete the following information):

• Version [1.2.2]
• OS: [DeepinOS]

Size of butt/ass, (maybe) colour, flat/firm/athletic-muscled ass. :)

ubuntu@ubuntu-VirtualBox:~/Documentos/dreampower-master$ python3 main.py checkpoints download
[INFO] Downloading https://cdn.dreamnet.tech/releases/checkpoints/v0.0.1.zip
[INFO] Extracting /tmp/tmp6bebckp4/v0.0.1.zip
[ERROR] stat: path should be string, bytes, os.PathLike or integer, not dict
[ERROR] Something Gone Bad Download Downloading The Checkpoints

Why does my dreamtime only use v0.0.1 dreampower?Is this normal?

Is it possible to train the model? I mean i cant see any options for training. How can i help training the model?

F:\dreampower>python main.py -i 123.jpg --ignore-size
[INFO] Welcome to DreamPower
[INFO] GAN Processing Will Use GPU IDs: [0]
[WARNING] Image Size Requirements Unchecked.
[INFO] Executing Image Transform
[INFO] Processing on 123.jpg
[INFO] Executing Dress To Correct
[INFO] Executing Correct To Mask
[INFO] Executing Mask To Maskref
Traceback (most recent call last):
File "main.py", line 144, in
argv.ArgvParser.run()
File "F:\dreampower\argv.py", line 373, in run
args.func(args)
File "F:\dreampower\main.py", line 33, in main
select_processing().run()
File "F:\dreampower\processing_init_.py", line 40, in run
self.execute()
File "F:\dreampower\processing_init.py", line 163, in _execute
r = p.run(*[self._image_steps[i] for i in p.input_index])
File "F:\dreampower\transform_init.py", line 29, in run
r = self.execute(*args)
File "F:\dreampower\transform\opencv\mask.py", line 47, in execute
res1 = cv2.bitwise_and(correct, correct, mask=green_mask_inv)
cv2.error: OpenCV(4.1.0) C:\projects\opencv-python\opencv\modules\core\src\arithm.cpp:245: error: (-215:Assertion failed) (mtype == CV_8U || mtype == CV_8S) && _mask.sameSize(*psrc1) in function 'cv::binary_op'

Building package on dreampower directory (Archlinux) after checkpoints:

makepkg -si
Installing pyinstaller...
ERROR: Can not perform a '--user' install. User site-packages are not visible in this virtualenv.

any idea to solve this issue?
TY

If the --ignore-size parameter is given, the algorithm is run without requiring the 512x512 input image size.

(using this with a non 512x512 image will result in bad result images currently, it's a step towards being able to accept non 512x512 images in the future)

As the title, can not start with win 10 from v1.2.0 to v1.2.2. But Ok with v1.1.0 at the same PC.Do I need any other programs ?

For me the current release almost always generates weird pixels on vag area, usually green/yellow colored. Using default parameters. It never happens for boobs. I'm already using split colored background. Happens for square images too.
Is it because of low quality jpg saving at one of the stages?

"Dreampower checkpoints download" seems non functional on Windows 10.

Installed version 1.0.0, had to find the checkpoints manually.

Describe the bug
When using the color correction option, the subject of the transformation will be corrected to match the color of the removed garments. For example, the skin of a person wearing a red shirt would turn red after the transformation.

To Reproduce
Steps to reproduce the behavior:

1. Turn on color correction
2. Perform a transformation on an image with a colorful shirt
3. See error

Expected behavior
The skin tone to match the original image when using color correction

Screenshots

Enviroment (please complete the following information):

• Dreamtime 1.2.1
• OS: Windows 10 1903

Additional context
This is just conjecture and I haven't actually examined how the algorithm functions, perhaps this issue could be fixed via setting the parts of the image identified as clothes to transparent in both the before and after images before feeding them into the color correction algorithm, then using the difference in hue to modify the original output,

Can’t reach 100% CPU load neither with these: --n-cores=2, 4, 8
Even selected 8cores only goes up to 53% CPU load peak (33% average).
No SSD operation is happening at that time.

Which means wasting time by not fully utilizing the CPU.
Is there a bottleneck in the code?

To Reproduce
Simple run: --auto-resize --n-cores=4 (no CUDA)

Enviroment (please complete the following information):
PC: intel i5 Windows 10 x64 (2cores, 4 logical CPUs)
Python 3.6.7
latest SW checkpoint