Image segmentation is the task of labeling an image at the pixel level, and it underpins much of modern medical imaging, autonomous perception, and scene understanding. Here we break down segmentation architectures, loss functions, and evaluation practices, with frequent attention to the medical settings where boundary accuracy and robustness matter most. Each piece traces the method back to its source research.
A team from NTU and Fudan University identified a blind spot that has haunted referring expression AI for a decade: the assumption that every phrase points to exactly one object. Their fix is a new family of benchmarks, a 259K-expression…
Researchers at Central South University fine-tuned the Segment Anything Model 2 with full-parameter adaptation, a cross-scale feature fusion decoder, and a hybrid loss function — building a single model that achieves up to 98.13% mIoU across 13 material microscopy datasets…
SAMM: SAM2 Fine-Tuned for Universal Material Micrograph Segmentation Read More »
Land cover mapping from aerial and satellite imagery has always been treated as a deterministic task by default. A convolutional network takes in a patch, produces a class distribution for every pixel, and outputs one hard prediction map. What it…
A bayesian Segmentation Model That Flags Its Own Uncertain Pixels Read More »
Most medical segmentation systems are trained to find one thing. A polyp. An organ. A lesion. Nobody explicitly teaches the network what the tissue around that target actually looks like. The model learns foreground boundaries entirely from positive examples, so…
PraNet-V2 Fixes Medical Segmentation By Modeling Background Read More »
Researchers at Chongqing University of Technology built a u-shaped encoder-decoder that fuses dynamic sparse attention from BiFormer with a redesigned channel-spatial skip connection — outperforming TransUNet by 4.49% DSC and Swin-Unet by 3.34% DSC on the Synapse multi-organ benchmark while…
BRAU-Net++: U-Shaped Hybrid CNN-Transformer Network for Medical Image Segmentation Read More »
Researchers at Sichuan University of Science and Engineering built a network that fuses reverse attention, multi-scale criss-cross self-attention, and a Laplacian edge filter borrowed from classical image processing to delineate colorectal polyp boundaries with a precision and speed that most…
Researchers at Hohai University built a network that treats high-frequency spectral components — the sharp edges, fine contours, and subtle discontinuities that standard convolutions quietly destroy — as first-class semantic cues rather than noise to be filtered away, and the…
FreDNet: The Remote Sensing Segmenter That Learned to Hear the Image, Not Just See It Read More »
Segmentation models correct each other’s mistakes through a “tight coupling” phenomenon – which makes them brutally hard to fool in a black‑box setting. Researchers discovered that by destroying global semantic integrity and remapping local features, transfer attack success jumps by…
A compact temporal attention module called TAM quietly outperforms much heavier architectures on cardiac segmentation — by making existing networks motion-aware without rebuilding them from scratch.
TAM: Plug-and-Play Temporal Attention Module for Motion-Guided Cardiac Segmentation Read More »
Optical coherence tomography angiography arrived in clinical practice with a peculiar contradiction. It captures volumetric blood flow at micron resolution, yet the vast majority of analysis tools treat it as a stack of 2D images. The enface projection — a…
MT-Net: 3D Retinal Microvascular Segmentation via Multi-Scale Topology Regulation Read More »