ResNeSt: Split-Attention Networks
This work addresses the need for more efficient and accurate neural networks for computer vision tasks, offering a novel approach that enhances performance in image classification and transfer learning, though it is incremental in building upon existing attention and multi-path methods.
The paper tackles the problem of improving visual recognition by introducing ResNeSt, a modularized architecture that combines channel-wise attention with multi-path representation, resulting in a model that outperforms EfficientNet in accuracy-latency trade-off on image classification and achieves superior transfer learning results on public benchmarks.
It is well known that featuremap attention and multi-path representation are important for visual recognition. In this paper, we present a modularized architecture, which applies the channel-wise attention on different network branches to leverage their success in capturing cross-feature interactions and learning diverse representations. Our design results in a simple and unified computation block, which can be parameterized using only a few variables. Our model, named ResNeSt, outperforms EfficientNet in accuracy and latency trade-off on image classification. In addition, ResNeSt has achieved superior transfer learning results on several public benchmarks serving as the backbone, and has been adopted by the winning entries of COCO-LVIS challenge. The source code for complete system and pretrained models are publicly available.