From GANs to Diffusion: GDA for Perception Tasks | HackerNoon

Table of Links

Abstract and 1 Introduction

2. Related work

   2.1. Generative Data Augmentation

   2.2. Active Learning and Data Analysis

3. Preliminary

4. Our method

   4.1. Estimation of Contribution in the Ideal Scenario

   4.2. Batched Streaming Generative Active Learning

5. Experiments and 5.1. Offline Setting

   5.2. Online Setting

6. Conclusion, Broader Impact, and References

A. Implementation Details

B. More ablations

C. Discussion

D. Visualization

2. Related work

2.1. Generative Data Augmentation

Generative data augmentation (GDA) refers to using generative models to synthesize additional data for augmentation. With the continuous improvement in the capabilities of generative models (Goodfellow et al., 2020; Saharia et al., 2022; Rombach et al., 2022b), GDA has become a popular technique for improving model performance. Several works use GDA in perceptual tasks such as classification (Feng et al., 2023; Zhang et al., 2023; Azizi et al., 2023), detection (Zhao et al., 2023; Chen et al., 2023), and segmentation (Li et al., 2023b; Wu et al., 2023b;a; Xie et al., 2023). Early works (Zhang et al., 2021; Li et al., 2022) involve the use of generative models like Generative Adversarial Networks (GANs) (Goodfellow et al., 2020) to generate additional training data. With the evolution of diffusion models, recent works (Azizi et al., 2023; Li et al., 2023b; Wu et al., 2023b; Yang et al., 2023; Zhao et al., 2023) favor using high-quality diffusion models such as Imagen (Saharia et al., 2022) and Stable Diffusion (Rombach et al., 2022b) for data generation. X-Paste (Zhao et al., 2023) has proven the strategy of using copy-paste to be more effective than directly using generated data for mixed training, and for the first time demonstrated that using generated data can enhance the performance of segmentation models on the long-tailed segmentation dataset LVIS (Gupta et al., 2019). Therefore, we consider it as the baseline for our work. However, while previous works have investigated the effects of GDA on different tasks, there has been limited exploration on how to better filter and utilize generative data for downstream perception models.

:::info
Authors:

(1) Muzhi Zhu, with equal contribution from Zhejiang University, China;

(2) Chengxiang Fan, with equal contribution from Zhejiang University, China;

(3) Hao Chen, Zhejiang University, China ([email protected]);

(4) Yang Liu, Zhejiang University, China;

(5) Weian Mao, Zhejiang University, China and The University of Adelaide, Australia;

(6) Xiaogang Xu, Zhejiang University, China;

(7) Chunhua Shen, Zhejiang University, China ([email protected]).

:::

:::info
This paper is available on arxiv under CC BY-NC-ND 4.0 Deed (Attribution-Noncommercial-Noderivs 4.0 International) license.

:::