@AlessandroFlabo

Co-founder of Procedo | PhD in Computer Science @SapienzaRoma | Member of the Perception and Intelligence Lab. @SapienzaPINlab

 CTO and Founder of Procedo
procederai.com
Joined November 2012
  • Tweets3.8K
  • Following518
  • Followers367
🆕News about our oral work at #IROS2024 The final version of the paper is available! arxiv.org/abs/2407.13567 We've released the code, check it (and leave a⭐️): github.com/GDam90/hyp2nav See you in Abu Dhabi! @GuidoDamely @PascalMettes @GalassoFab10
Fabio Galasso@GalassoFab1028 Jul 2024
Selected as oral presentation at #IROS 2024: Hyperbolic geometry endows reinforcement learning with uncertainty awareness, for novel robot route-planning policies that identify possible collisions in crowds. ArXiv: arxiv.org/abs/2407.13567 Code to be released soon

Hyp2Nav: Hyperbolic Planning and Curiosity for Crowd Navigation

Autonomous robots are increasingly becoming a strong fixture in social environments. Effective crowd navigation requires not only safe yet fast planning, but should also enable interpretability...

arxiv.org
3
1
10
668
We introduce COSKAD, a model that improves anomaly detection (AD) by using skeletal motion data instead of raw video frames. We explore three latent spaces for skeletal embeddings: Euclidean, hyperbolic, and spherical.
1
2
83
Our work "Contracting Skeletal Kinematics for Human-Related Video Anomaly Detection" published in the Pattern Recognition journal! Paper: sciencedirect.com/science/ar… Code: github.com/aleflabo/COSKAD Made in @SapienzaPINlab (@GuidoDamely @stdrr_ @AlessioSampier1 @GalassoFab10 )
1
2
1
5
570
POV: You're at the #CVPR2024 concert and you really like the music, but don't have a banner 🚀
5
33
6,289
When they ask you which is the best #CVPR2024 paper: P.s. yes, our paper is called "PREGO: Online Mistake Detection in PRocedural EGOcentric Videos" and you can find it here openaccess.thecvf.com/conten… See you in Seattle 🚀
1
11
541
📢 Our #CVPR2024 is finally out! "PREGO: online mistake detection in PRocedural EGOcentric videos" We introduce PREGO, a model for detecting procedural errors in egocentric videos as they occur. Paper: arxiv.org/abs/2404.01933 Code: github.com/aleflabo/PREGO (available soon)
1
4
4
19
3,561
Dilly ding dilly ding 🔔 We are in #CVPR2024 🚀 "PREGO: online mistake detection in PRocedural EGOcentric Videos" has been accepted to @CVPR 🍷 Impressive effort from the whole team! @GalassoFab10, @GuidoDamely, @LeonardoPlini, @lu_scofano, @edmatteis, @anfurnari, @GMFarinella
1
6
29
1,970
Here is the PINlab team taking a break from the #CVPR2024 deadline 🥵 – at Rome, Lazio
1
2
23
1,036
We're presenting our work in room nord poster 139! We have some stickers too 😏 #ICCV2023
Alessandro Flaborea@AlessandroFlabo22 Aug 2023
🔥 Exciting News! 🔥 Our paper "Multimodal Motion Conditioned Diffusion Model for Skeleton-based Video Anomaly Detection" has been accepted to the #ICCV2023 conference! Paper: arxiv.org/abs/2307.07205 Code: github.com/aleflabo/MoCoDAD Here I describe some insights into our work:
Overview of the proposed MoCoDAD. A sequence of N skeletal motions is split into past (top-right X1:k frames) and future (top-left Xk+1:N frames). During training, the Forward Diffusion block adds noise to the future frames, shifting each joint by a random vector displacement of varying intensity (increasing with the diffusion timestep t). Then the Reverse Diffusion learns to estimate the noise. A key aspect of MoCoDAD is the conditioning, i.e. how to encode the past clean k frames and guide the synthesis of relevant futures.

ALT Overview of the proposed MoCoDAD. A sequence of N skeletal motions is split into past (top-right X1:k frames) and future (top-left Xk+1:N frames). During training, the Forward Diffusion block adds noise to the future frames, shifting each joint by a random vector displacement of varying intensity (increasing with the diffusion timestep t). Then the Reverse Diffusion learns to estimate the noise. A key aspect of MoCoDAD is the conditioning, i.e. how to encode the past clean k frames and guide the synthesis of relevant futures.

7
350
Paris, I'm coming! See you @ #ICCV2023
7
240
Load more