YOGA: Deep object detection in the wild with lightweight feature learning and multiscale attention. (July 2023)

• Record Type: Journal Article
• Authors: Sunkara, Raja; Luo, Tie
• Is Part Of: Pattern recognition. Volume 139(2023)
• Publication Date: 2023-07
• Subjects: Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2023.109451
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: YOGA is a new object detection model that learns richer representation via attention based multi-scale feature fusion with a much lighter model that reduces nearly half convolution filters. We provide a theoretical explanation of how label smoothing facilitates backpropagation during training, by mathematically analyzing how the loss gradient vector is involved in the recursive backpropagation algorithm. We also overcome overfitting using Genetic Algorithm based hyper-parameter tuning. We compare YOGA with over 10 state-of-the-art deep learning object detectors and demonstrate the superiority of YOGA on the joint performance of model size and accuracy. We migrate YOGA to real hardware (Jetson Nano 2GB) to assess its usability in the wild. Our experiments show that YOGA is well suited for even the lowest-end deep learning edge devices. Abstract: We introduce YOGA, a deep learning based yet lightweight object detection model that can operate on low-end edge devices while still achieving competitive accuracy. The YOGA architecture consists of a two-phase feature learning pipeline with a cheap linear transformation, which learns feature maps using only half of the convolution filters required by conventional convolutional neural networks. In addition, it performs multi-scale feature fusion in its neck using an attention mechanism instead of the naive concatenation used by conventional detectors. YOGA is a flexible model that can be easily scaled up or down by several orders of magnitude to fit a broad range of hardware constraints. We evaluate YOGA on COCO-val and COCO-testdev datasets with over 10 state-of-the-art object detectors. The results show that YOGA strikes the best trade-off between model size and accuracy (up to 22% increase of AP and 23–34% reduction of parameters and FLOPs), making it an ideal choice for deployment in the wild on low-end edge devices. This is further affirmed by our hardware implementation and evaluation on NVIDIA Jetson Nano.