Jinkun Cao will present his great work at #ICRA2021 today at 9pm GMT+1. He did this work while visiting me at UC Berkeley as an undergraduate student. It is definitely my pleasure to see that things went well and he is a Ph.D. student at CMU now.

We are hosting Workshop on Autonomous Driving again this year #CVPR2025 2021. It features a list of amazing speakers as well as challenges on Waymo Open Dataset and Argoverse! Happening on June 20! cvpr2021.wad.vision

Jiangmiao Pang did this great work when visiting us at UC Berkeley. We are all excited about the amazing results and method simplicity. You can hear more from us at 11:00 AM – 1:30 PM in New York time at #CVPR2025 today! More at vis.xyz/pub/qdtrack

Thanks to Alex Liniger and Yuning Chai, the videos of the keynote talks at the Workshop on Autonomous Driving has been well organized on Youtube youtube.com/channel/UCNnC_…

📢 I am excited to announce that BDD100K model zoo is open for business! It contains more than 100 pre-trained models for various tasks. Each model also comes with results and visualization on val and test sets. You can also put your model in the zoo! 👉github.com/SysCV/bdd100k-…

In this ICCV2021 week, I will discuss how we can better solve the basic perception problems for autonomous driving in 3 workshops AVVision Workshop (avvision.xyz/iccv21/#speake…) 3D Object Detection from Images（sites.google.com/unitn.it/3dodi) ROAD Challenge (sites.google.com/view/roadchall…)

📢📢📢Human pose annotation is available on BDD100K now! 14K frames and 27K persons are labeled with 18 joints for each person. We also released pre-trained models in the BDD100K model zoo. You can also test your model on our evaluation server. Links👇

📢Lei Ke is presenting PCAN, Prototypical Cross-Attention Networks for Multiple Object Tracking and Segmentation at NeurIPS 2021 as Spotlight. PCAN uses test-time prototypes to memorize instance appearance and achieve impressive seg tracking accuracy on YT and BDD100K. Links👇

Our BDD100K (bdd100k.com) tracking challenges at CVPR 2022 are now live! We hope our challenges can lead to new advances in computer vision and new pragmatic algorithms on self-driving cars. Challenge details: go.yf.io/bdd100k-cvpr22

Mask Transfiner, appearing #CVPR2025 2022, produces high-resolution instance segmentation with extreme details without high computation or memory cost. This is enabled by a new principal mechanism called Incoherent Region Prediction. More at vis.xyz/pub/transfiner/

It is my great pleasure to announce a new release of Scalabel, an open-sourced software that provides a versatile and scalable annotation platform, supporting both 2D and 3D data labeling. Link: github.com/scalabel/scala… 🧵

The world is continuously changing, but the deep learning models have a hard time keeping up. We are presenting a synthetic dataset called SHIFT for continuous multi-task domain adaptation at #CVPR2022 on Fri pm at 257B. Poster, video, more at vis.xyz/shift

Uncertainty is critical for neural networks to guess what they don't know. Deterministic methods promise strong performance with negligible cost, or do they? We did a reality check. Janis Postels will present our findings tomorrow at #ICML2022. more at vis.xyz/pub/deu-practi…

BDD100K challenges at ECCV 2022 are going on! Similar to CVPR 2022, our main challenges are on MOT and MOTS, but with much bigger prizes! Top winners will get $5000! We are also opening two teaser challenges to track without tracking labels. More @ bdd100k.com/challenges/ecc…

As #ECCV2022 European Conference on Computer Vision #ECCV2026 is happening this week, I hope you get a chance to enjoy the nice weather and beautiful beach at Tel Aviv. This afternoon at local time, I will give an invited talk on 4D scene understanding for self-driving at AVVision Workshop (avvision.xyz/eccv22).

Cameras on self-driving cars can see in every direction. @sy2es93104 will present our new work CC-3DT for 3D object tracking on the panoramic views #CoRL2022. CC-3DT makes the gap between vision-only and lidar-based 3D object tracking much smaller. More at vis.xyz/pub/cc-3dt

Segment Anything is fun and powerful but still requires full mask annotation for supervised training. Our new work MaskFreeVIS appearing at #CVPR2023 shows we can get accurate video instance segmentation WITHOUT any video or image mask annotations. More at vis.xyz/pub/maskfreevis

ONLY 4 weeks until the submission deadline to BDD100K challenges at #CVPR2023! Time to show off your multi-object tracking (and seg) methods! The winners will get cash awards and a podium to present your work at one of the largest workshops at #CVPR! More: vis.xyz/bdd100k/challe…

Monocular depth estimation is “impossible” because one image can’t measure depth geometrically. Our iDisc #CVPR2023 can group pixels w/o supervision and learn depth inductive bias on groups. We get LiDAR-like (but denser) depth from single images! More: vis.xyz/pub/idisc

We are at #CVPR2023 to present our research works. Please check our poster and talk schedule at vis.xyz/news/2023-06-17. Feel free to come to say hi!