@davnords.bsky.social
Phd Student @ Chalmers Deep Learning for Computer Vision. Strengthen your ViTs: https://github.com/davnords/octic-vits
23.10.2025 05:35 — 👍 1 🔁 0 💬 0 📌 0
Nope, don't think so. Just inward facing I think, e.g. CO3D and WildRGBD
22.10.2025 23:08 — 👍 1 🔁 0 💬 0 📌 0Turns out NLP is just vision
21.10.2025 16:39 — 👍 24 🔁 5 💬 2 📌 2
These gentlemen show how not only colmap but also VGGT fail on spherical motion in their ICCV oral paper "Uncalibrated Structure from Motion on a Sphere".
I wonder if it is just a data issue for VGGT or if it is deeper than that. I mean VGGT was trained on mostly synthetic data. What do you think?
Cursed RoPE
16.10.2025 14:56 — 👍 3 🔁 0 💬 0 📌 0
Pro tip: For good Halloween vibes, use non-normalized RoPE on images larger than your training resolution and larger than the composite period of some of the RoPE-rotations. You might get scary ghost structures in your features.
16.10.2025 14:53 — 👍 11 🔁 3 💬 1 📌 1Maybe not directly comparable, but in some sense the academic vision community has not scaled up so tremendously
12.10.2025 20:36 — 👍 0 🔁 0 💬 0 📌 0Certainly a big leap. Another perspective I sometimes entertain is that AlexNet used around 60 million trainable parameters and when reading conference papers nowadays, many of the main experiments are on models of around ViT-B size which has around 85 million parameters.
12.10.2025 20:35 — 👍 0 🔁 0 💬 1 📌 0GigaSund>GigaChad
05.10.2025 14:07 — 👍 1 🔁 0 💬 0 📌 0Big if true ;)
01.10.2025 06:22 — 👍 0 🔁 0 💬 0 📌 0Never told us if accuracy or loss: Sudden drop = Nuke the Berzelius cluster
30.09.2025 18:09 — 👍 0 🔁 0 💬 1 📌 0Saw a post on "The bitter lesson's bitter lesson"
28.09.2025 13:00 — 👍 3 🔁 0 💬 1 📌 0A. Andersson sounds like the most Swedish generic name I could imagine 😆
26.09.2025 09:29 — 👍 3 🔁 0 💬 1 📌 0We'll take it 😂
26.09.2025 09:28 — 👍 0 🔁 0 💬 0 📌 0Fun fact: this is not artificial lighting. The literal grace of god shone upon you
26.09.2025 07:14 — 👍 1 🔁 0 💬 0 📌 014 opponents? 'Kids these days have it too easy', doctorate inflation
25.09.2025 09:54 — 👍 0 🔁 0 💬 0 📌 0
RoMa now on PyPI under name of `romatch`
23.09.2025 19:56 — 👍 7 🔁 2 💬 1 📌 0goated map
18.09.2025 11:26 — 👍 2 🔁 0 💬 0 📌 0
Towards the Next Generation of 3D Reconstruction
@parskatt.bsky.social PhD Thesis.
tl;dr: would be useful in teaching image matching - nice explanations. (too) Fancy and stylish notation. Cool Ack section and cover image.
liu.diva-portal.org/smash/record...
And here is a link to the thesis itself: liu.diva-portal.org/smash/record...
17.09.2025 06:31 — 👍 19 🔁 6 💬 0 📌 0
How to name your method: a comprehensive flow chart
13.09.2025 15:32 — 👍 43 🔁 10 💬 1 📌 0Live esport events are legendary!
06.09.2025 17:45 — 👍 1 🔁 0 💬 0 📌 0A wild Horace He appears :)
26.08.2025 15:51 — 👍 1 🔁 0 💬 1 📌 0Reasonable take
25.08.2025 08:53 — 👍 1 🔁 0 💬 1 📌 0Interesting. Do you know if this is unique to DINO-style pre-training or is it present in all early ViT layers?
24.08.2025 09:42 — 👍 1 🔁 0 💬 1 📌 0Next we need a dataset of octic-symmetric octupi, or seals :)
21.08.2025 16:02 — 👍 2 🔁 0 💬 0 📌 0Thanks for the vote of confidence. Then you two are lazily-efficient!
17.08.2025 17:54 — 👍 1 🔁 0 💬 0 📌 0Rotate conference venues between the likes of Hawaii and Japan and we will have infinite travel glitch
17.08.2025 17:53 — 👍 2 🔁 0 💬 0 📌 0I like the "try to implement X from scratch" where X is some frontier paper. Though, I don't always take the time to do it because I am lazy
16.08.2025 12:46 — 👍 2 🔁 0 💬 1 📌 0
