BestHub
Sign in
AIWalker
AIWalker
Aug 18, 2025 · Artificial Intelligence

UniConvNet: Expanding Effective Receptive Field for a SOTA CNN Vision Backbone (ICCV 2025)

UniConvNet introduces a three‑layer receptive‑field aggregator that combines small kernels to enlarge the effective receptive field while preserving its Gaussian distribution, achieving state‑of‑the‑art results on ImageNet‑1K, COCO2017 and ADE20K with only 30M parameters and 5.1G FLOPs.

CNNEffective Receptive FieldICCV2025
0 likes · 6 min read
UniConvNet: Expanding Effective Receptive Field for a SOTA CNN Vision Backbone (ICCV 2025)
AI Frontier Lectures
AI Frontier Lectures
Apr 4, 2025 · Artificial Intelligence

How OverLoCK Redefines Vision Backbones with Dynamic Convolution

OverLoCK, a new vision backbone inspired by human top‑down attention, combines a three‑stage decomposition, dynamic ContMix convolutions and top‑down guidance to achieve state‑of‑the‑art performance on ImageNet classification, COCO detection and ADE20K segmentation while maintaining strong trade‑offs.

OverLoCKTop-down AttentionVision Backbone
0 likes · 10 min read
How OverLoCK Redefines Vision Backbones with Dynamic Convolution