ModelScope--Cartoonization of portraits, skin beautification of portraits

ModelScope

What is ModelScope? ModelScope is an open-source model-as-a-service sharing platform that provides pan-AI developers with flexible, easy-to-use, and low-cost one-stop model service products, making model application easier! Here you can experience the training model for free, and at the same time achieve model prediction with one line of commands, quickly verify the model effect, and at the same time, you can customize your own personalized model, and you can also improve your model research and development capabilities while experiencing it, and this is very lively. Community, the road to growth is not alone. Other ModelScope introductions and usage Welcome to the ModelScope Documentation Center, which has all the answers you want.

Here you can first take a look at the overview of ModelScope and the services that ModelScope can provide us. Given the services it provides, what businesses can we implement. Below I will mainly demonstrate the powerful functions of ModelScope through the demonstration of the cartoon model of the portrait.

model library

Since we want to demonstrate the cartoonized portrait model, we need to use the model library first, and find the cartoonized portrait model in the model library.

The Dharma cartoon model here is actually what we commonly call a portrait cartoon model. There is also a line of red small characters in the picture to identify [Portrait Cartoonization]. After experiencing the portrait cartoonization, I will demonstrate the [Portrait Skin Beauty Model] later.

Dharma cartoon model

Next, we will demonstrate the Dharma cartoon model. On the model introduction page, we can also see the animation of the cartoonized portrait display effect that has been realized, as well as the description of the model, the method and scope of use, and how to use it.

On the right side of the model introduction, you can directly upload local pictures for testing. Here is the effect of my test, which is good. In addition, what I am demonstrating is to open the model file online through Notebook for testing.

Portrait beauty model

After demonstrating the cartoonized model of Dharma, I will demonstrate the [Portrait Skin Beautifying Model]. Why do I mainly demonstrate these two models? First of all, I am more interested in these two models. Of course, it is also because everyone is also interested in these two models. I like it more, after all, skin beautification and cartoonization are good solutions to make photos beautiful. The introduction page of portrait skin model model.

Compared with the original picture, the test results of the online test are really whitened, haha

Environmental preparation

You can choose two ways to develop, one is to develop through the local environment,
One is online development through Notebook, here I choose online development with Notebook, why? Because you see that you need to install the Python environment, the deep learning framework, and the ModelScope Library locally, it is too complicated, and you should run it directly in the Notebook. The official image provided in the Notebook does not need to install the environment independently, which is more convenient and fast. It is recommended that you also try it. try.

Notebook development

Open the notebook login page, if you don't have an account, you need to register one.

Experience record

Document usage record

In the experience process, basically all operations need the help of documentation. After all, this is a new business for me personally, and the fastest way to understand and accept the new business is to read the documentation and try it out. The documentation as a whole is aimed at The introduction of ModelScope, model operation, notebook operation guide, model library, data set, developer guide, etc. are basically perfect. The only missing part should be the Q&A part of the problems in use, because some problems are not experienced by experiencers. It may be a common problem caused by the operation, but the document does not list the solution to this kind of problem, which leads to the problem that cannot be dealt with in time during the use process, which affects the efficiency and affects the user experience.

expected result

Let me put it this way, although I encountered a little problem in the process of experiencing the portrait skin beautifying model, the expected effect has been achieved, and the overall experience is still good; about the expected effect, I hope that the processing of low-pixel pictures can be improved. The result is that the high-pixel image processing effect is better, and the adaptation of the low-pixel image model is not very good, and the image processing is not very satisfactory.
Applicable scene

Although I have only experienced two models of portrait cartoonization and portrait skin beautifying, I am very optimistic about the use of ModelScope. Just talking about the portrait cartoon model I experienced, there are still many young generations who like to take their own photos. Cartoonization is used to make avatars on various platforms, and now there are many apps or websites or third-party tools that can cartoonize pictures. Either the cartoonization effect of portraits is not good, or they have to be charged, or they have watermarks. Really Whoever uses it is blocked, so when it comes to the cartoonization of ModelScope's portrait, the effect is really much better than the outside.

Product linkage

As mentioned above, ModelScope's portrait cartoon effect is good, then whether it can provide an interface to a third-party app or website, and the app or website provides a function called portrait cartoonization, which directly calls the interface of ModelScope to achieve cartoonization effect, just talking about portrait cartoon Personally, I don’t think a company needs it, but any product that uses user avatars has the possibility of being connected, and it can be said that the prospect is very good.

Summarize

Although ModelScope has not been used before, the overall feeling of this experience is still very good, although I have tried many times when I experienced the portrait skin beauty model and kept reporting errors. After asking professionals, I switched to GPU mode to create a notebook development environment. After successfully achieving the experimental effect, there are still many experience models, such as English image description, Dharma portrait matting and many other models. The ModelScope model covers multi-modalities such as image and vision, and involves speech, natural language processing, computer vision, etc. It can be said that it involves all aspects of life, whether it is personal experience, personal use or commercial use. At the same time, through open community cooperation, build deep learning related models to open source, and open source related models to serve innovative technologies to promote the prosperity of the model application ecosystem. Development is really technology to change life!