We introduce a flexible memory extension mechanism, allowing users to adapt based on FPS, frame count, and other data characteristics. Our model is fast and lightweight, requiring minimal GPU memory. (6/7)

Even without bidirectional information flow (unlike offline models), our approach achieves state-of-the-art results among comparable online and offline tracking models across multiple datasets. (5/7)

Unlike traditional methods relying on full temporal modeling, our model operates causally—processing frames without future information. We introduce two memory modules: (i) Spatial Memory, addressing feature drift; (ii) Context Memory, storing full tracking history. (4/7)

We simply process points as queries in a transformer decoder. Instead of regressing coordinates (as in dominant methods), we treat tracking as a classification problem, selecting the most likely patch per query and refining with local offsets. (3/7)

Unlike prior work focused on offline point tracking, we target online tracking on a frame-by-frame basis, making it ideal for real-time, streaming scenarios. At the core of our approach is a simple yet effective transformer-based model. (2/7)

you mean like AGI is bad 🤔
today at Pera.

and very honest about the limitations:

Efe, one of our interns last summer, implemented the ROACH expert on NAVSIM and apparently shared the code: github.com/efemantarci/...

there is no documentation but it should be fairly easy to use in case you need an RL expert reaching the performance of TransFuser on NAVSIM.

Mısra will present Odd-One-Out next week at #ACCV2024, check out her talk in Oral Session 5 on day 3 (December 12 at 9:00).

kuis-ai.github.io/O1O/

unfortunately, I won't be there -teaching is falling apart due to too much traveling-, but I'm looking forward to visiting Vietnam in the future 🙂

vow Sadra won the best poster award at #BMVC 🥳
I should have paid for the gala dinner this time 😂

the funniest thing on the internet today.

here we go, heroes++
we had a good run bsky 👋

look at this?!

in a country where it rains and snows all the time, there was an hour of sunshine today during our visit to #GlasgowNecropolis.
one hour Scotland, I needed one hour of your default gloomy weather with rain and darkness. result: cheery photos in a gothic cemetery 🙄

Sadra and I will be in Glasgow for the next few days, presenting our work with Matteo and Fabio at #BMVC tomorrow. I will be visiting some friends in Edinburgh afterwards and give a talk on Friday, looking forward to it all! last time I visited Scotland was in August, I cannot imagine late Nov 😅