
PhD candidate @UofT / @VectorInst. Senior researcher @Waabi_ai 🚙 🚚 🚛
Toronto, Ontario
Joined November 2013
- Tweets66
- Following259
- Followers211
- Likes367
Pinned Tweet
How do we perform 3D object detection well for all object classes of interest, including rare classes in self-driving data? 👀🚗
In our #CoRL2025 paper, we introduce FOMO-3D, the first LiDAR-camera detector that uses 2D vision foundation models for closed-set 3D detection. (1/8)
Joyce Yang retweeted
Transluce is developing end-to-end interpretability approaches that directly train models to make predictions about AI behavior.
Today we introduce Predictive Concept Decoders (PCD), a new architecture that embodies this approach.
Joyce Yang retweeted
Can we reconstruct dynamic driving scenes without any labels? Check our #NeurIPS2025 paper Flux4D, a flow-based, generalizable, unsupervised 4D reconstruction model that scales to real-world driving data!
Website: waabi.ai/flux4d
Arxiv: arxiv.org/abs/2512.03210
[1/n]
Please read our paper tinyurl.com/fomo-3d for more details! If you are attending CoRL, please visit us at Poster #74 in Poster Session 1 this afternoon. (7/8)
1
2
188
Many thanks for @RubiksCubeMan79 who led the project with me, and our co-authors @Dvornik_Nikita, Thomas Li and @RaquelUrtasun! (8/8)
2
157
Joyce Yang retweeted
How do we explain the activation patterns of neurons in language models like Llama?
I'm excited to share work that we did at @TransluceAI to inexpensively generate high-quality neuron descriptions at scale!
Scaling Automatic Neuron Description
We trained open-weight models that automatically describe the patterns of neuron activations in language models, producing high-quality explanations on par with a human expert.
Full report: transluce.org/neuron-descrip…
Joyce Yang retweeted
Could we reconstruct large scenes in high quality in 30s? Yes! Excited to share our #ECCV24 paper, G3R: Gradient-Guided Generalizable Reconstruction (waabi.ai/g3r), which learns a network to iteratively refine the predicted 3D representation with 3D gradient feedback.
Joyce Yang retweeted
#CVPR2024 Workshop on AV Simulation!
Simulation is a crucial tool to accelerate the development of safe autonomous driving (AD).
In our full-day CVPR workshop with exciting speakers from Waymo, Tesla, Waabi, Wayve, Nuro, Princeton and many others, we’re bringing together 1/N
Joyce Yang retweeted
We are really excited about the release!! The dataset is live now with instructions and an example notebook. The sequences are specially selected to be very challenging! Please feel free to reach out to either Joyce or myself with any questions! 😃
📢📢📢 The AMV-Bench dataset is now released! This SLAM benchmark features 7 cameras (including a stereo pair) and 116 sequences of real-world driving with high-quality ground-truth, diverse environments, weather and motion paths. Check out bit.ly/amv-slam for more!
📢📢📢 The AMV-Bench dataset is now released! This SLAM benchmark features 7 cameras (including a stereo pair) and 116 sequences of real-world driving with high-quality ground-truth, diverse environments, weather and motion paths. Check out bit.ly/amv-slam for more!
Today at 4pm EDT we're presenting Asynchronous Multi-View SLAM at #ICRA2021 🤖🚗
Large-scale #SLAM for async multi-camera robots and a new 480km benchmark w/ 100+ sequences.
Who: @ajyang99, Can Cui, @RaquelUrtasun, @ShenlongWang
Web: https://nitter.cf/t.co/T5JPamnHoA 🧵⬇️ (1/5)
Joyce Yang retweeted
Can we simulate intricate lighting effects for driving scenes? Check our #NeurIPS paper LightSim which combines neural and physics-based rendering for realistic, controllable and scalable lighting simulation!
Website: waabi.ai/lightsim
Arxiv: arxiv.org/abs/2312.06654
[1/n]
Joyce Yang retweeted
"Trajeglish: Learning the Language of Driving Scenarios" w/ @xbpeng4 @FidlerSanja
Discrete sequence modeling for controlling interactive agents in self-driving simulation
@nvidia @VectorInst @UofTCompSci @SFU
research.nvidia.com/labs/tor…
arxiv.org/abs/2312.04535
1/6
Joyce Yang retweeted
Researchers from Waabi and the University of Toronto Introduce LabelFormer: An Efficient Transformer-Based AI Model to Refine Object Trajectories for Auto-Labelling
Quick read: marktechpost.com/2023/11/13/…
Paper: arxiv.org/abs/2311.01444
Project Page: waabi.ai/labelformer/
#ArtificialIntelligence #DataScience @Waabi_ai @ajyang99
How can we generate high-quality auto-labels for self-driving LiDAR data without a human in the loop? 🚗
Check our @corl_conf paper LabelFormer, a simple, efficient, effective transformer-based model to refine object trajectories for auto-labelling.
More: waabi.ai/labelformer
Please read our paper arxiv.org/abs/2311.01444 and visit the website waabi.ai/labelformer/ for more details! If you are attending CoRL, our poster is at 4:15pm ET tomorrow Nov 9. (5/6)
1
177
Many thanks to my co-authors @sergioksas @Dvornik_Nikita @seanseg @yuwen_x @Jordan_Hu @carter_fang and @RaquelUrtasun !! (6/6)
185












