Some time ago, I’ve worked on an oil sac (energy reservoir) detection in copepods (kind of zooplankton) for Université Laval. Apart from presenting the results on the ML4Plankton conference I’ve published a detailed blog post on the Appsilon blog that I strongly encourage you to read! ...
Recently I was working at work on detecting the oil sac in copepods. This is was a super cool problem in which we received data from professor Frédéric Maps from Université Laval. We together gave a talk on conference ML4Plankton that you can watch online! First Frederic gives some context on why this problem is important and later I explain methods used and problems you can find. Our talk is available online here: ...
This is the second post from micro-series preceding the SFI conference. In the first post I showed how easy it is to do the style transfer at home. Now, I would like to present you how to play with StyleGANv2. StyleGANv2 is the second version of StyleGAN, they are very similar in core principals. We will be working with StyleGANv2 but I will refer to it as SG. There is a lot that can be said/explained about SG. Here I would like to focus on showing how to make your own experiments with SG instead of getting into SG’s details. ...
I wrote another post for Appsilon, this time about deploying streamlit applications on RStudio Connect infrastructure. This post might be totally useless for majority of audience, but a pure gold for a few 💰. What is streamlit you can understand from my previous post. Having a running app in your browser is already cool, but how cool is it to deploy your app on the internet? If you need an enterprise solution for hosting dashboards in RShiny/streamlit, RSConnect is the way! In this blog post I focused on presenting solutions for common problems you might encounter. ...
As mentioned earlier, I will be giving a talk on the 17th SFI IT Academic Festival on the topic of style transfer and StyleGAN. In this post I would like to show you how easy it is to play with style transfer on your own! To tutorial or not to tutorial It might be obvious for you to check Tensorflow/PyTorch tutorials if you wanted to learn how to create style-transfer system. But what if you do not want to learn anything? Maybe all you want is just to do the style transfer on your photos? Apparently tutorials with colab notebooks are all you need! ...
I am happy to share that I was invited (for the first time in my life!) to give a talk on a conference. 🎉 It will be the 17th SFI IT Academic Festival, organized by students from the Jagiellonian University in Cracow. The conference will be held on 14-16 of March 2022. My lecture will be in Polish on topic StyleGAN w Twoich rękach -> StyleGAN in Your Hands, I will talk about transfer learning in general and later about StyleGANv2. The exact time and place will posted on SFI webpage. You can expect some materials regarding the talk later on my blog! ...
This one is short. :) Usually when we are doing ML experiments, training models, preparing EDA, we want to share the results with the World at some point. One of the quickest and easiest ways to do so in python, to create a simple webapp with results, is a super-cool library streamlit! If you thought building an app in flask is easy, you will be amazed by streamlit simplicity. Actually you do not have to create an ML related app, nobody checks that ;). ...
As mentioned a month ago, I write a series of posts on detecting solar panels from satellite images at my company, Appsilon. In this post I focus on using fastai python library to deliver a working ML model in few hours! ...
I’m writing a series of posts on detecting solar panels from satellite images at my company, Appsilon. To be honest, not from satellite images but from orthophotos. If you don’t know what are orthophotos, the good news is that the first post has been released recently!. ...
Recently I wrote a post about DVC at my company’s, Appsilon blog. DVC is like a git, but for data, models and experiments. It also allows for creating an automated experiments pipelines. As a teaser I’ll just say that, having prepared scripts for model training and evaluating, when new data is added to the repo, the whole training is run automatically. Metrics are saved to appropriate files alongside with parameters, same with plots. However, if you just upload a new test data, a part of pipeline will be run only. This is a pretty simple case, but DVC is a very powerful tool, check out different use cases in their docs. ...