Apple Maps 3DV
Senior Machine Learning Engineer
I build the models behind Apple's Detailed City Experience: segmentation, detection and diffusion over street imagery, and pixel-to-vectorized-3D-map models.
Apr 2024 - Present
Work
Results
Third Wave Automation
Head of Perception and Manipulation
I led Perception and Manipulation on Third Wave's autonomous forklift platform, from pallet detection and pose estimation to barcode scanning and the vehicle's state machine.
Jun 2020 - Mar 2024
What Can You Learn from Your Muscles? Learning Visual Representation from Human Interactions
Kiana Ehsani, Daniel Gordon, Thomas Nguyen + 2 more
ICLR 2021
A visual representation learned from unlabeled human interaction video, where gaze and IMUs provide the labels as opposed to human-made annotations to learn human-centric representations.
Mar 2021
Learning by Watching and Learning by Doing
Daniel Gordon
Ph.D. Thesis 2020
My PhD thesis, on learning visual representations from videos and embodied agents.
Aug 2020
Watching the World Go By: Representation Learning from Unlabeled Videos
Daniel Gordon, Kiana Ehsani, Dieter Fox + 1 more
arXiv 2020
Videos provide a free source of semantically meaningful temporal labels, and Video Noise Contrastive Estimation uses that signal to learn robust image representations that outperform supervised ImageNet pretraining with similar dataset sizes.
Mar 2020
Introduction to Deep Learning
Ali Farhadi, Daniel Gordon, Aaron Walsman + 8 more
I was head TA for the University of Washington's Introduction to Deep Learning course, creating homeworks in a from-scratch Pytorch-autograd-style neural network library pure Python, NumPy, and Numba.
Sep 2019 - Dec 2019
ALFRED: A Benchmark for Interpreting Grounded Instructions for Everyday Tasks
Mohit Shridhar, Jesse Thomason, Daniel Gordon + 5 more
CVPR 2020
ALFRED maps natural-language instructions and egocentric vision to action sequences for household tasks, from 25k expert demonstrations in interactive environments.
Dec 2019
SplitNet: Sim2Sim and Task2Task Transfer for Embodied Visual Navigation
Daniel Gordon, Abhishek Kadian, Devi Parikh + 2 more
ICCV 2019
SplitNet decouples perception from policy learning, transferring between simulators and generalizing to unseen environments faster than end-to-end baselines.
May 2019
Shifting the Baseline: Single Modality Performance on Visual Navigation & QA
Jesse Thomason, Daniel Gordon, Yonatan Bisk
NAACL 2019
Unimodal baselines beat the random and majority-class ones typically reported, by up to 29% absolute, and are the right comparison for multimodal claims.
Apr 2019
What Should I Do Now? Marrying Reinforcement Learning and Symbolic Planning
Daniel Gordon, Dieter Fox, Ali Farhadi
arXiv 2018
HIP-RL merges hierarchical symbolic planning with deep reinforcement learning, reaching state-of-the-art results on three datasets with fewer test-time steps.
Dec 2018
Learning to Perform a Tetris with Deep Reinforcement Learning
Daniel Gordon
2018
An RL agent trained end-to-end to play Tetris discovers super-human policies, planning many moves ahead rather than greedily clearing what is available.
Jun 2018
Re3: Real-Time Recurrent Regression Networks for Visual Tracking of Generic Objects
Daniel Gordon, Ali Farhadi, Dieter Fox
RA-L 2018
We develop a real-time generic deep object tracker that uses an LSTM to update its appearance model on the fly in one forward pass, tracking at 150 FPS.
May 2018
IQA: Visual Question Answering in Interactive Environments
Daniel Gordon, Aniruddha Kembhavi, Mohammad Rastegari + 3 more
CVPR 2018
IQA asks questions that require both seeing and interacting with a scene. We develop the Hierarchical Interactive Memory Network to answer these questions with controllers at multiple levels of abstraction.
May 2018
AI2-THOR: An Interactive 3D Environment for Visual AI
Eric Kolve, Roozbeh Mottaghi, Daniel Gordon + 9 more
AI2 TR 2017
AI2-THOR is a framework of procedurally generated synthetic 3D indoor scenes where agents navigate and interact, spanning RL, planning, question answering and detection.
Dec 2017
Visual Semantic Planning using Deep Successor Representations
Daniel Gordon*, Yuke Zhu*, Eric Kolve + 5 more
ICCV 2017
We develop the deep successor representation for decoupling perception and action in order to learn end-to-end policies that transfer to unseen items and environments.
Jul 2017
Deep Learning in Practice
Daniel Gordon
A seminar on applying modern deep learning to real research problems, including the unglamorous parts that decide whether a model ships.
Jun 2017
Collaborative Imaging of Urban Forest Dynamics: Augmenting Rephotography to Visualize Changes over Time
Ruth West, Abby Halley, Jarlath O'Neil-Dunne + 2 more
IS&T/SPIE 2013
Augmented rephotography combined with LiDAR monitors the urban tree canopy, a community-based way to measure and visualize ecological change over time.
Feb 2013
Collaborative Rephotography
Ruth West, Abby Halley, Daniel Gordon + 2 more
SIGGRAPH 2013
Collaborative rephotography lets many people revisit many sites over time, from nation-scale scenic overlooks to NYC street trees.
Jan 2013
Nothing matches those filters.
Try removing a topic.