PeopleofSuperintelligence

Builder

Francisco Massa

Also known as Francisco Vitor Suzano Massa

Francisco Massa builds software and models for computer vision, from PyTorch tools and object detectors to self-supervised visual systems. His early work connected images with 3D models, and he helped start PyTorch-Ignite while co-authoring the PyTorch paper. He co-authored Detectron2 and DETR, and the Detectron software team received the 2021 PAMI Everingham Prize. DINOv2 and DINOv3 extend his work to self-supervised visual features. He works at Meta.

Timeline

  1. 2014–2017

    Linking images and 3D models

    His 2014 study with Mathieu Aubry and Renaud Marlet used convolutional neural networks for object detection and 3D pose estimation in still images. His 2017 thesis is titled Relating images and 3D models with convolutional neural networks.

  2. 2017–2019

    Building with PyTorch

    He started PyTorch-Ignite with Alykhan Tejani and Joost van Amersfoort at the end of 2017. PyTorch's 2019 paper, which he co-authored, describes an imperative, Pythonic deep-learning library.

  3. 2019

    Extending object detection tools

    He co-authored Detectron2, Facebook AI Research's successor to Detectron and maskrcnn-benchmark. The platform provides object-detection, segmentation, and visual-recognition tools.

  4. 2020

    Predicting objects as sets

    With Nicolas Carion and four other co-authors, he develops DETR, which treats object detection as direct set prediction. Its Transformer encoder-decoder predicts objects in parallel and extends to panoptic segmentation.

  5. 2021

    A prize for Detectron

    In 2021, the Detectron object-detection and segmentation software team received the PAMI Everingham Prize at ICCV. Francisco Massa was one of its members.

  6. 2023

    Learning visual features

    He co-authors DINOv2, a self-supervised method that learns robust visual features from curated image data. The team trains a billion-parameter Vision Transformer and distills it into smaller models.

  7. 2025

    Scaling self-supervised vision

    He co-authors DINOv3, a self-supervised model for visual representations across natural and aerial images. Its Gram anchoring method helps address dense-feature degradation during long training schedules.

  8. As of October 2026

    Working at Meta

    He works at Meta after completing a PhD with Renaud Marlet.

Sources and credits

Photo: We have no published photo of Francisco Massa yet. Send us a free-licensed photo.

  1. [1412.7190] Convolutional Neural Networks for joint object detection and pose estimation: A comparative study arxiv.org. Checked 14 October 2026.
  2. Katalog - SLUB Dresden katalog.slub-dresden.de. Checked 14 October 2026.
  3. PyTorch: An Imperative Style, High-Performance Deep Learning Library proceedings.neurips.cc. Checked 14 October 2026.
  4. About us — PyTorch-Ignite v0.4.3 Documentation docs.pytorch.org. Checked 14 October 2026.
  5. GitHub - facebookresearch/detectron2: Detectron2 is a platform for object detection, segmentation and other visual recognition tasks. · GitHub github.com. Checked 14 October 2026.
  6. [2005.12872] End-to-End Object Detection with Transformers arxiv.org. Checked 14 October 2026.
  7. ICCV 2021 PAPER AWARDS | ICCV 2021 iccv2021.thecvf.com. Checked 14 October 2026.
  8. [2304.07193] DINOv2: Learning Robust Visual Features without Supervision arxiv.org. Checked 14 October 2026.
  9. [2508.10104] DINOv3 arxiv.org. Checked 14 October 2026.
  10. Mathieu Aubry mathieuaubry.github.io. Checked 14 October 2026.