Projects
Segment Your Ring (SYR)
Using Segment Anything Model (SAM) released by Meta, the objective is to segment rings used in product packshots. The challenging part is that SAM doesn’t perform well with rings on a white background. Therefore, I adpated SAM with LoRA to solve this task.
Inpainting with Stable Diffusion and Segment Anything
Stable diffusion and segment anything are two foundation models that were recently released. I combined stable diffusion inpainting model that requires an image and a mask with segment anything that generates masks (with prompts or automatically) to custom pictures. The project has a webui made with gradio.
Cat face generator
GAN models are generative models that train a generator through a discriminator in a zero sum game. I tried to generate cats face with a DCGAN model.
Tweet virality predictor
The popularity of a tweet can vary depending on the author, the audiance and other parameters. In this project, I built a full pipeline from getting tweets to predict the virality of them with a Hawkes process fine-tuned by a RandomForest. Finally I deployed the system with Docker and Kubernetes.
Convex contouring obstacles on water with Kernel Density Estimation
This project calculate the convex contour of probability given a cloud of points.
Evaluation of marine ecosystem through plankton species
Plankton are a good estimator of marine ecosystem. In the context of the Biodiversity AI Challenge, I was asked to create a model that classify plankton species based on a highly unbalanced dataset.