Francisco Massa
Computer Vision Research Engineer at Facebook
Paris, Ile-de-France
Summary
🤩
Rockstar🎓
Top SchoolFrancisco Massa is a Computer Vision Research Engineer based in Paris with 11 years of experience building and optimizing vision models and GPU-accelerated libraries. He has a strong academic foundation (PhD in Computer Vision and Machine Learning) and has applied that research rigor to production-scale projects at Facebook and Twitter. His open-source footprint spans core deep learning tooling—contributions to PyTorch, torchvision, and xformers—and influential vision repos like DeiT and maskrcnn-benchmark, where he implemented distillation, sparse/memory-efficient attention, and CUDA optimizations. Francisco combines low-level CUDA/kernel work and backend fixes with high-level model engineering, improving performance, correctness, and distributed training primitives. Comfortable moving between research and engineering, he has a track record of fixing subtle numerical and memory issues while shipping practical features used by the community.
11 years of coding experience
1 year of employment as a software developer
Doctor of Philosophy (PhD), Computer Vision and Machine Learning, Doctor of Philosophy (PhD), Computer Vision and Machine Learning at Ecole Nationale des Ponts et ChausséesMaster 2 (M2), Machine Learning, Master 2 (M2), Machine Learning at ENS CachanBachelor of Science (B.Sc.), Engenharia Elétrica e Eletrônica, Bachelor of Science (B.Sc.), Engenharia Elétrica e Eletrônica at Universidade Estadual de CampinasIngénieur polytechnicien, Electrical Engineering, Ingénieur polytechnicien, Electrical Engineering at Ecole polytechnique
Portuguese, French, English, German
Github Skills (45)
unit-testing10
pytorch10
c-language10
dis10
operation10
python10
attention-mechanism10
tensorrt10
sparse-matrix10
gpu-programming10
machine-learning10
distill10
c1110
c1710
lua10
Programming languages (14)
Github contributions (5)
facebookresearch/maskrcnn-benchmark
Oct 2018 - Oct 2019
Fast, modular reference implementation of Instance Segmentation and Object Detection algorithms in PyTorch.
Role in this project:
ML Engineer
ML EngineerContributions:1 release, 23 commits, 197 PRs in 11 months
Contributions summary:Francisco primarily contributed to the Mask R-CNN benchmark by implementing new features and fixing existing issues. Their contributions included adding support for Python 2 compatibility, which involved modifying import statements and handling differences in library functions. They also addressed weight initialization in a predictor, and added the option to postprocess masks during inference, enhancing the model's functionality. The user also worked on updating and adding configurations for the Pascal VOC dataset.
pytorchdeep-learningobject-detectioncomputer-visionmodular
facebookresearch/xformers
Nov 2021 - Dec 2022
Hackable and optimized Transformers building blocks, supporting a composable construction.
Role in this project:ML Engineer
ML EngineerContributions:258 reviews, 93 commits, 48 PRs in 1 year 1 month
Contributions summary:Francisco primarily focused on developing and optimizing components for a Transformer-based machine learning library. Their contributions included introducing and refining the `SparseCSRTensor` and `BlockSparseTensor` data structures to enhance sparse tensor operations. The user also worked on improving the memory-efficient attention mechanism, with both forward and backward passes implemented in CUDA, along with several optimizations for improved performance. Additionally, they enhanced the library's integration with PyTorch's dispatcher.
pytorchtransformersbuilding-blockshackableoptimized
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.