Jump to content

Upsample and Stylize images with Machine Learning on Vectorworks Cloud Services

Recommended Posts

  • Vectorworks, Inc Employee

Hi everyone,

The Vectorworks Cloud Services team is very excited to announce two Machine Learning-based features:


Stylize images: Vectorworks Service Select members can use Vectorworks Cloud Services to create an artistic design from plain renderings, making them more interesting with almost no effort or time.


Upsample images: Vectorworks Service Select members can create a low resolution rendering in Vectorworks, and then use Vectorworks Cloud Services to upsample the rendering, to get a good-quality rendering in less time.


Here are the simple steps you can take to try these features:


Step 1:

Sign in to https://cloud.vectorworks.net.


Step 2:

Right click on an image file that is uploaded to your Vectorworks Cloud Services account, or drag and drop into the window to upload,  and select Stylize image or Upsample image. Works for panorama images as well!

Screen Shot 2020-03-18 at 2.48.18 PM.png.


Step 3:  

In the dialog box, select how and where to save the output image, choose one of the predefined styles to apply to your image (for Stylize image), or the scale (for Upsample image) and then select Done to get the cloud job submitted.


Step 4:

To see a detailed status, select the Status button Screen Shot 2020-03-18 at 2.51.57 PM.png at the top right of the page.


Please try them out and let us know if you like them or how we can improve them.


Have fun!


The Vectorworks Cloud Services team



  • Like 2
Link to comment

I don't get this? 

Why would you use this, when affinity , photoshop or pixelmator pro are quicker, better quality and have more options for tone mapping/stylising? 

- quick test of upscaling:  an image in pixelmator pro took about 6 seconds, vectorworks cloud took over 5 minutes and was a smaller image? 

Am I missing something? 

Link to comment
  • Vectorworks, Inc Employee

If you already have those apps then please go ahead and use them.


As these ML features become more integrated with Vectorworks itself (and they will over time) it may save you some time and steps to use these built-in rather than piping to other applications.  For example we plan to add upsampling of renderings and animations ASAP, and every frame will benefit.


  • Like 1
Link to comment
Guest Ben Atwood

Traditional image upscaling, like in Photoshop or any other graphics program, is fundamentally different from machine-learned image upsampling.  If your graphics program is producing results that work for you, please keep using it!



This new feature increases the resolution of an image and uses machine learning to ADD detail to the new image that wasn't there in the original smaller image.  Traditional scaling creates a larger resolution image with no extra detail, and uses something like cubic interpolation to fill in the extra pixels.  This results in a blurry large image, which may be better than a pixelated image.


For some comparisons, here is an image where each pixel is 4x its original size:

Training Room Zoomed Crop.png


Here is the image upscaled in GIMP using cubic interpolation.  Note that the pixelation is gone but the image is blurry.

Training Room Scaled Crop.png


And here is the ML upsampled version.

Training Room Upsampled Crop.png


The sharp lines remain sharp, the fine textures are retained, and the text is more readable.  This technique does introduce artifacts of its own and some parts are blurry in places where you might want them to be sharp.


It is essentially impossible to train the neural network to do this perfectly, and yes it is much slower than using Photoshop.  However, we hope the results may be much better for some applications, like panoramas and producing images for large-scale prints.  If a particular rendering is already pushing the limits of what your computer (or the cloud) can produce in a reasonable amount of time, but you still need a higher-resolution image, we hope this feature will help.  A few minutes per image might seem like a lot compared to your graphics program, but consider that re-rendering at 4x the resolution will take roughly 16x longer than the original rendering.


As Dave said, we hope to integrate this with our other rendering workflows in the future to produce high-resolution results in a shorter time.

Link to comment
  • 1 month later...



I was very excited about the stylize image feature, but am disappointed that it limits me to only a few styles. Hoped I could choose my own example, because that's how it came across in the design summit. Any future plans for an enhancement?



Link to comment

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

  • Create New...