I wanted to create a model that can identify the name of the composer from a music. But that’s long way out. For now, I wanted to revector the bird v. forest to identify the composers from various images downloaded from ddg.
I’d downloaded pictures of them with a nuance of ‘portrait’ and ‘group’ pictures. I thought this will download Mozart by himself and in a crowd. I didn’t think much that there were no cameras in Beethoven and Mozart’s era
So, I was concerned to see the show_batch() included pictures of orchestra playing Mozart
I just finished lesson 1 of the course, and thought I would share my simple yet tasty types of ramen classification model (miso, shio, shoyu, and tonkotsu). Just impressed with how easy fastai library makes it to download images, train, and test the model.
Please see below link to the classification model, which performs pretty well after just 3 epochs of fine tuning. Looking forward to learning and sharing more!
Trying Out document classification with Vision model
I’m excited to share my progress on the Fast.ai course! At my current organization, we were contemplating using Amazon Comprehend for a document classification use case. However, I decided to take a different approach and embarked on a project to build a document classifier using a vision model, although I’m still evaluating if this is the optimal strategy. Initially, I generated two styles of PDFs , converted those to images using a script generated from ChatGPT, and then fine-tuned a ResNet model using these document images. The resulting model is deployed on Hugging Face. My future plans for this project involve data extraction and storage in a database, and I’m considering creating templates for each PDF to facilitate data extraction. I’m open to any suggestions or insights from the community on how to further develop this project. I’d also like to express my gratitude to Jeremy for this incredible course, which has been truly inspiring. As of now, I’m progressing through the lessons on gradient boosting
Hello, fastai community :). Excited to finally get started on this course.
For lesson 1 & 2 I made a recognizer of street art styles. In Paris where I live, there are many street artists active and often it’s hard to tell which piece of work was done by whom. The recognizer is trained on example works of 4 artists (50 samples from each).
this summer ib Berlin a huge police search was going on after a video went viral that supposedly showed a lioness at night. After some fearful days the authorities decided that is probably was a boar after all. I built an image classifier that might have spared them this wild goose chase. Find it here: BigCat Or Boar - a Hugging Face Space by ChrZeller
After five rounds of fine tuning, the error rate on the validation set was a surprisingly low 1.7% - even though the dataset was imperfect! Call me pleasantly surprised. Definitely looking forward to learn and experiment more with future lessons.
Thanks Jeremy for this wonderful course! I am beginning my journey with it. This is my first work - classifying buildings/structure design so that we can know if it is based on ancient or modern design. My_Notebook
Good morning everyone!
My friend and I recently wrote a book about practices in setting up RAG, based on our experience trying out LangChain, llamaindex, haystack etc.
I’d like to share it with ML/AI enthusiasts here.
I’m really happy that I’ve managed to deploy the model that I toyed with in week one in week 2. I retrained it and cleaned the data set, but it still was quite small. Probably going to come back to this sometime when I have more experience on building a data set, not just duckduckgo-ing photos.
It correctly identified my zinnias with 51% confidence and 43% being sure they were daisies, which I’ll count as a win.