Predicting multiplex immunofluorescence marker expression from H&E whole slide images using ViT foundation models.
Predicting multiplex immunofluorescence marker expression directly from H&E whole slide images using ViT foundation models. Published in Computers in Biology and Medicine (2026).
Winning Solution of Task 2 MIDOG 2025 (MICCAI 2025) - We finetune DINOv3-H+ model with LoRA for atypical mitotic figure classification across scanner and tissue domains
Fine-tuning DINOv3 pretrained on natural images for atypical mitotic figure classification across scanner and tissue domains. Winning Solution.
Implementation of A-LAMP for photo aesthetic quality assessment on the AVA dataset.
End-of-studies project (2019–2020) at Télécom SudParis, in collaboration with Meero. Implementation of Shuang Ma et al. A-LAMP paper for aesthetic quality scoring on the AVA dataset. With Fabiola Espinoza and Simon Albergel.
Citation: Ma, S., Liu, J., & Wen Chen, C. (2017). A-lamp: Adaptive layout-aware multi-patch deep convolutional neural network for photo aesthetic assessment. In Proceedings of the IEEE conference on computer vision and pattern recognition
Predicting wealth levels across African regions from satellite imagery and nighttime light data.
2nd-year project at Télécom SudParis (2018–2019). Predicting wealth levels across African regions using satellite imagery and nighttime light intensity data. Explored CNN transfer learning, Gaussian process regression, and classical regression methods. This project was inspired by this paper
Citation: Chitturi, Varun, and Zaid Nabulsi. “Predicting poverty level from satellite imagery using deep neural networks.” arXiv preprint arXiv:2112.00011 (2021).
Image is from: here
U-Net with EfficientNet-B4 for pneumothorax segmentation on chest X-rays; ranked 98/1475.
Kaggle SIIM-ACR Pneumothorax Segmentation competition (summer 2019). Built a U-Net with EfficientNet-B4 encoder and ResNet decoder to segment pneumothorax regions in chest X-rays. Ranked 98/1475 (top 7%, bronze medal) with a Dice score of 0.8407 on the private leaderboard.
Detecting and tracking players and referees in football match videos with possession metrics.
Internship project between 1st and 2nd year at Télécom SudParis (summer 2018). Detecting and tracking players and referees in football match videos to compute possession metrics and distance traveled. Because data was scarse, we used captured videos from FIFA 18. Built with OpenCV, YOLO, TensorFlow.
Real-time face detection, tracking, and identification from video.
Big Brother is a student project at Télécom SudParis to build a full facial recognition system from scratch.
It performs face detection, tracking, and identification from live video using methods like MTCNN, dlib, and FaceNet, with matching based on feature similarity.
The project also highlights the capabilities and risks of facial recognition technologies.