So I tested it out and it doesn't seem to work
hey @eatcats thanks for your interest in this, the problem here is that we trained our model on just one pre–made dataset provided by a researcher a while back then. And so the model has only one imagination of the dataset on which it has been trained on. If you are really interested and want to fine–tune this, I can help you with that but you'd need your own dataset with lot of variations in it, like the way you created this input mock–up image.
Hi, can the model be trained on dribbble image designs to work?
@ElyseNova777
Ahm yes as soon as you also provide the related html and css code into the dataset, then you have to tokenize it and for each design or parts of the design you need to repeat these steps until you've created a ton of dataset and then you need to annotate and train the model.
The challenging process of creating such a Model is the collection of the dataset, which is very time-consuming if you want to train your model with high-quality designs and high-quality HTML, CSS. Another way of creating such a big Dataset is by collecting synthetic Data, but with synthetic Data, you won't get the high-quality results of individual complex designs like the dribbble designs. Training an AI Model depends on what you put in... you put garbage in, you'll get garbage out. A solid Dataset is all that really matters
@taneemishere I am interested in fine tuning this model with my own dataset of designs, can you help?
@ElyseNova777 I'd really like to answer your question but i don't think so I can add more into the problem the way @w4lk3rm00n has explained. At all, you'd need your own dataset plus the html and css code for it, that is really time consuming and hectic process. This work was done on top of a pre–made dataset, and I only done the modelling part while I was doing my thesis to show the potential. btw thanks for the explanation @w4lk3rm00n
@absg this again depends upon how much data you have and how organised that is, the mockups the html css code, plus other such things described in above comments. Also I can help but I can't guarantee you cause I'm busy in my full time job and other research work, and only got Sundays free and sometime Saturdays. But the code is available on my GitHub you can see and replicate, fine tune and train your own model with your data. Also if you do so, just let me know so that I can link to your models as well if it produces great results.