ScaleNAS: Multi-Path One-Shot NAS for Scale-Aware High-Resolution Representation

Hsin-Pai Cheng, Feng Liang, Meng Li, Bowen Cheng, Feng Yan, Hai Li, Vikas Chandra, Yiran Chen

Abstract: Scale variance among different sizes of body parts and objects is a challenging problem for visual recognition tasks. Existing works usually design dedicated backbone or apply Neural architecture Search (NAS) for each task to tackle this challenge. However, existing works impose significant limitations on the design or search space. To solve these problems, we present ScaleNAS, a one-shot learning method for exploring scale-aware representations. ScaleNAS solves multiple tasks at a time by searching multi-scale feature aggregation. ScaleNAS adopts a flexible search space that allows an arbitrary number of blocks and cross-scale feature fusions. To cope with the high search cost incurred by the flexible space, ScaleNAS employs one-shot learning for multi-scale supernet driven by grouped sampling and evolutionary search. Without further retraining, ScaleNet can be directly deployed for different visual recognition tasks with superior performance. We use ScaleNAS to create high-resolution models for two different tasks, ScaleNet-P for human pose estimation and ScaleNet-S for semantic segmentation. ScaleNet-P and ScaleNet-S outperform existing manually crafted and NAS-based methods in both tasks. When applying ScaleNet-P to bottom-up human pose estimation, it surpasses the state-of-the-art HigherHRNet. In particular, ScaleNet-P4 achieves 71.6% AP on COCO test-dev, achieving new state-of-the-art result.

Keywords: automatic machine learning, neural architecture search, evolutionary search

One-sentence Summary: The authors present a one-shot based search method and novel search space for multi-scale neural architectures

Track: Main track

Reproducibility Checklist: Yes

Broader Impact Statement: Yes

Paper Availability And License: Yes

Code Of Conduct: Yes

Reviewers: Hsin-Pai Cheng ([e-mail])

Steps For Environmental Footprint Reduction During Development: Fine-tuning the performance of Neural Architecture Search crafted model is computation consuming which may cause potential environmental impact (e.g., carbon footprint and global warming). Our work vastly reduces the design and evaluation time and brings positive environmental impact by enabling a more flexible design space with the proposed efficient one-shot searching method. Without further fine-tuning, our searched architectures can be directly deployed onto devices.

CPU Hours: 900

GPU Hours: 900

TPU Hours: 0

Evaluation Metrics: Yes

Estimated CO2e Footprint: 126

Class Of Approaches: Evolutionary search, one-shot NAS

Datasets And Benchmarks: MPII, COCO, CrowdPose

Performance Metrics: mIoU, mAP (mean Average Precision)

Benchmark Performance:

Multi-Person Pose Estimation COCO test-dev, 71.6 AP
Multi-Person Pose Estimation CrowdPose, 71.3 mAP @0.5:0.95

Benchmark Time:

Multi-Person Pose Estimation COCO test-dev, 3 GPU days
Multi-Person Pose Estimation CrowdPose, 3 GPU days