Attention mechanisms, vision transformers, and the architectures replacing convolutions across vision tasks. We unpack how attention is used, misused, and reinvented in current research, from efficient attention branches to Mamba-style state space models.
Modern perception systems are increasingly built by bolting pretrained models together rather than training one big model end to end. You take a 3D model that understands language, add an image model that can segment on command, wire them with…
Foundation Model: Why Collapsing to One Class Too Early Loses Information Read More »
Show an object detector a picture of an elephant standing in a road and it might tell you, with high confidence, that it is looking at a car. This is not a rare glitch. Detectors are trained on a fixed…
SAM3-O2D2 Catches Unknown Objects by Prompting a Foundation Model Read More »
Point a camera at a stick insect on a branch and your eye does most of the work without you noticing. A neural network does not have that luxury. When a target shares the same color and texture as everything…
LGFN Fuses RGB and Polarization to Spot Camouflage Read More »
Pair a normal camera with a thermal one and you get a detector that should work day or night. That is the promise of RGB and thermal fusion. The catch is that neither camera is trustworthy all the time. Drive…
RA-SOD Detects Salient Objects When Sensors Fail Read More »
Watch two people shake hands at a reception while a third pours a drink nearby, and your eye never loses track of whose hand is whose. A detector does. Most hand detectors find hands one at a time as separate…
Hand Object: One Query That Links Both Hands to a Person Read More »
Before an oncologist chooses a cancer drug, they are making an educated guess. Will this patient’s tumour respond, or will the treatment burn precious weeks while the cancer keeps growing? The truest answer would come from watching, inside the patient’s…
XPert: A Transformer That Predicts How Drugs Change Cells Read More »
Vision transformers earned their place in medical imaging because self attention lets a model connect a finding in one part of an image to context somewhere else entirely, the kind of long range reasoning a small convolution kernel cannot do…
FadeFormer: Graph Diffusion Sharpens Medical Image Classification Read More »
Vision transformer surveys are common enough that a new one needs a clear organizing idea to stand out. This paper leans on a memorable framing, four secrets, and mostly earns it. Rather than organizing around a chronological list of named…
The Four Secrets Behind Video Vision Transformers Explained Read More »
Attention became the default tool for sequence and spatial reasoning for a good reason. It lets every element look at every other element and decide, dynamically, what matters. That flexibility is also its cost. Comparing every pair of positions scales…
How Mamba State Space Models Are Reshaping Multimodal AI Fusion Read More »
Transformers rewrote what was possible in language and vision, so pointing them at graphs looked like a sure bet. The bet mostly paid off on small graphs and mostly stalled on large ones. The reason is simple to state and…
A Graph Transformer That Scales to Billions of Nodes Read More »