Andrei Bârsan (@andreib) header
Andrei Bârsan (@andreib) profile picture

Robotics researcher @waabi_ai | PhD candidate at the University of Toronto | Past: Uber ATG Toronto | Opinions are my own, etc.

Adamstown, Pitcairn Islands
Joined April 2009
Photo by Andrei Bârsan (@andreib)Photo by Andrei Bârsan (@andreib)Photo by Andrei Bârsan (@andreib)Photo by Andrei Bârsan (@andreib)Photo by Andrei Bârsan (@andreib)Photo by Andrei Bârsan (@andreib)Photo by Andrei Bârsan (@andreib)Photo by Andrei Bârsan (@andreib)Photo by Andrei Bârsan (@andreib)
Andrei Bârsan retweeted
Can we build 3D digital twins scalably from sensor data captured by a moving platform and use them for realistic and controllable simulation? Check our paper CADSim openreview.net/forum?id=Mp3Y…, presenting tmr at #CoRL2022 (Spotlight & Poster: 9:35-10:50 AM, OGG-FPAA & Room 040B-C)!
2
12
57
7,570
Check this out! How to solve inverse problems in an efficient and scalable manner while leveraging the observations to give feedback to the network.
Tired of hand-crafting energy functions for your inverse problems? Bored of deep nets predicting estimations that are not coherent with the observations? Deep feedback inverse problem solver to the rescue! It’s fast, robust, and generic! #ECCV2020 #Spotlight
4
21
Very exciting work, showing our new interpretable neural motion planner that does not require detection, and thus is safer as it avoids thresholding. Check it out at #ECCV2020
Can we learn safe #SelfDriving end-to-end while keeping interpretability? Yes! Check out our #ECCV2020 paper on neural #motionplanning through interpretable semantic representations. Joint work w/ A.Sadat, @mengyer, P.Dhawan, X.Wu, @RaquelUrtasun @UberATG arxiv.org/pdf/2008.05930.pdf
1
7
49
Introducing LevelSet R-CNN, a deep variational method for instance segmentation. We fuse classical optimization w/ modern deep learning, resulting in an elegant deep structured model. Paper: https://nitter.cf/t.co/4sRKCxvNwB (code coming!) w/ @namdarh,@yuwen_x,@JustinJWLiang,@weichiuma
2
15
62
Check out our paper "PolyTransform: Deep Polygon Transformer for Instance Segmentation" at #CVPR2020 [4PM-6PM 6/17 and 4AM-6AM 6/18 (PST)]. Joint work w/ @JustinJWLiang ,@namdarh ,@weichiuma ,@yuwen_xiong_cs ,Rui Hu Paper: https://nitter.cf/t.co/K3osU8ttQn CVPR: https://nitter.cf/t.co/3dx9xmispU
Boomers: “Oh no how will I avoid a patrolling robot the paths surely aren’t predictable how will I get to my objective” Millennials: “While you were out frolicking, I was in your basement, studying the blade”
17
44
1
452
Andrei Bârsan retweeted
I didn't have a good sense of how efficient the software FFMPEG libraries were, so my mind had already plotted various system optimizations, but it turns out that decoding 128 DVD streams simultaneously and scaling + RGB conversion only uses about 10% of the CPU on my 3990.
24
42
4
676
Andrei Bârsan retweeted
Who needs self-driving cars? nitter.cf/reformedbroker/status/…
14
301
57
1,045
Andrei Bârsan retweeted
We are organizing the 5th MOTChallenge workshop at CVPR'20 in Seattle, multi-object tracking and segmentation edition! We are going pixel-precise this time, hosting three exciting challenges! More info at motchallenge.net/workshops/b… @cvpr2020 @computervision
12
34
Andrei Bârsan retweeted
Each Thursday at OpenAI is Learning Day: a day where employees have the option to self-study technical skills that will make them better at their job but which aren't being learned from daily work. Here's how it works: openai.com/blog/learning-day…
23
322
58
1,287
Vector is proud to be partnered with @THP_hospital for the Machine Learning for Health (ML4H) Unconference, which focuses on connecting doctors and machine learning researchers to help define important healthcare opportunities that machine learning can help in solving. #ML4Health
11
40
Andrei Bârsan retweeted
EfficientNets: a family of more efficient & accurate image classification models. Found by architecture search and scaled up by one weird trick. Link: arxiv.org/abs/1905.11946 Github: https://nitter.cf/t.co/vrRnVJVQNh Blog: https://nitter.cf/t.co/0pPZmTxfUs
24
642
40
1,889
Andrei Bârsan retweeted
OpenAI Five makes history by winning a best-of-three versus @OGesports, the Dota 2 world champions!!!! Huge congrats to the team and to OG for an extremely well-played match.
26
187
44
653
Andrei Bârsan retweeted
Need to clean the LaTeX code of your research paper before submitting to arXiv? Use the arXiv LaTeX cleaner (goo.gl/ncLPZP), an open source tool by Google Zürich Research Scientist @jponttuset that automatically resizes images, erases unused images and removes comments.
Introducing arXiv LaTeX cleaner, an open source tool to make it safer and easier to open source research papers. The tool scrubs LaTeX code, deletes, and resizes images. We hope taking the pain out of the process encourages more to embrace open access! goo.gl/MXvcBP
1
211
14
652
Andrei Bârsan retweeted
Different parts of the brain learns in different ways. Model-free, model-based, and memory-based are all just different sides of the same coin, according to Kenji Doya #cosyne19
8
179
12
561
Andrei Bârsan retweeted
"What are the 3 biggest open problems in NLP?" We had asked experts a few simple but big questions for the NLP session at the @DeepIndaba. We're now happy to share the full responses from Yoshua Bengio, @redpony, @RichardSocher and many others docs.google.com/document/d/1…
13
441
28
983
Andrei Bârsan retweeted
GANs learn about eyes very early in training, and immediately begin sticking them everywhere. I'd like to think this one would make a good album cover.
14
91
24
468
Andrei Bârsan retweeted
Boring Co is launching a whole product line of DIY watchtowers. You get bricks & a picture.
304
1,555
131
27,031
Hahahahahahaha @nest found a way to make Bluetooth worse
9
40
1
227