Data-Driven Neuron Allocation for Scale Aggregation Networks
This work addresses the inefficiency in neuron allocation for multi-scale aggregation in deep networks, offering a method to enhance performance in visual recognition tasks like image classification and object detection, though it is incremental as it builds on existing architectures like ResNet.
The paper tackles the problem of manually allocating neurons for multi-scale information aggregation in visual recognition networks, which leads to suboptimal performance, by proposing a data-driven neuron allocation method that adaptively preserves the most informative neurons across scales, resulting in improved accuracy with low computational cost. On ImageNet classification, it reduces top-1 error rates by 1.12% (101 layers) and 1.82% (50 layers) compared to ResNets, and on COCO object detection, it improves mmAP by 3.6 (101 layers) and 4.6 (50 layers) with Faster RCNN.
Successful visual recognition networks benefit from aggregating information spanning from a wide range of scales. Previous research has investigated information fusion of connected layers or multiple branches in a block, seeking to strengthen the power of multi-scale representations. Despite their great successes, existing practices often allocate the neurons for each scale manually, and keep the same ratio in all aggregation blocks of an entire network, rendering suboptimal performance. In this paper, we propose to learn the neuron allocation for aggregating multi-scale information in different building blocks of a deep network. The most informative output neurons in each block are preserved while others are discarded, and thus neurons for multiple scales are competitively and adaptively allocated. Our scale aggregation network (ScaleNet) is constructed by repeating a scale aggregation (SA) block that concatenates feature maps at a wide range of scales. Feature maps for each scale are generated by a stack of downsampling, convolution and upsampling operations. The data-driven neuron allocation and SA block achieve strong representational power at the cost of considerably low computational complexity. The proposed ScaleNet, by replacing all 3x3 convolutions in ResNet with our SA blocks, achieves better performance than ResNet and its outstanding variants like ResNeXt and SE-ResNet, in the same computational complexity. On ImageNet classification, ScaleNets absolutely reduce the top-1 error rate of ResNets by 1.12 (101 layers) and 1.82 (50 layers). On COCO object detection, ScaleNets absolutely improve the mmAP with backbone of ResNets by 3.6 (101 layers) and 4.6 (50 layers) on Faster RCNN, respectively. Code and models are released at https://github.com/Eli-YiLi/ScaleNet.