The Unmet Promise of Synthetic Training Images: Using Retrieved Real Images Performs Better

• URL: http://arxiv.org/abs/2406.05184v2
• Date: Wed, 3 Jul 2024 06:00:50 GMT
• Title: The Unmet Promise of Synthetic Training Images: Using Retrieved Real Images Performs Better
• Authors: Scott Geng, Cheng-Yu Hsieh, Vivek Ramanujan, Matthew Wallingford, Chun-Liang Li, Pang Wei Koh, Ranjay Krishna,
• Abstract summary: Every synthetic image ultimately originates from the upstream data used to train the generator. We compare finetuning on task-relevant, targeted synthetic data generated by Stable Diffusion against finetuning on targeted real images retrieved directly from LAION-2B. Our analysis suggests that this underperformance is partially due to generator artifacts and inaccurate task-relevant visual details in the synthetic images.
• Score: 39.57368843211441
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Generative text-to-image models enable us to synthesize unlimited amounts of images in a controllable manner, spurring many recent efforts to train vision models with synthetic data. However, every synthetic image ultimately originates from the upstream data used to train the generator. What additional value does the intermediate generator provide over directly training on relevant parts of the upstream data? Grounding this question in the setting of image classification,a we compare finetuning on task-relevant, targeted synthetic data generated by Stable Diffusion -- a generative model trained on the LAION-2B dataset -- against finetuning on targeted real images retrieved directly from LAION-2B. We show that while synthetic data can benefit some downstream tasks, it is universally matched or outperformed by real data from our simple retrieval baseline. Our analysis suggests that this underperformance is partially due to generator artifacts and inaccurate task-relevant visual details in the synthetic images. Overall, we argue that retrieval is a critical baseline to consider when training with synthetic data -- a baseline that current methods do not yet surpass. We release code, data, and models at https://github.com/scottgeng00/unmet-promise.

Related papers

• DataDream: Few-shot Guided Dataset Generation [90.09164461462365]
  We propose a framework for synthesizing classification datasets that more faithfully represents the real data distribution. DataDream fine-tunes LoRA weights for the image generation model on the few real images before generating the training data using the adapted model. We then fine-tune LoRA weights for CLIP using the synthetic data to improve downstream image classification over previous approaches on a large variety of datasets.
  arXiv  Detail & Related papers  (2024-07-15T17:10:31Z)
• Is Synthetic Image Useful for Transfer Learning? An Investigation into Data Generation, Volume, and Utilization [62.157627519792946]
  We introduce a novel framework called bridged transfer, which initially employs synthetic images for fine-tuning a pre-trained model to improve its transferability. We propose dataset style inversion strategy to improve the stylistic alignment between synthetic and real images. Our proposed methods are evaluated across 10 different datasets and 5 distinct models, demonstrating consistent improvements.
  arXiv  Detail & Related papers  (2024-03-28T22:25:05Z)
• Improving the Effectiveness of Deep Generative Data [5.856292656853396]
  Training a model on purely synthetic images for downstream image processing tasks results in an undesired performance drop compared to training on real data. We propose a new taxonomy to describe factors contributing to this commonly observed phenomenon and investigate it on the popular CIFAR-10 dataset. Our method outperforms baselines on downstream classification tasks both in case of training on synthetic only (Synthetic-to-Real) and training on a mix of real and synthetic data.
  arXiv  Detail & Related papers  (2023-11-07T12:57:58Z)
• UAV-Sim: NeRF-based Synthetic Data Generation for UAV-based Perception [62.71374902455154]
  We leverage recent advancements in neural rendering to improve static and dynamic novelview UAV-based image rendering. We demonstrate a considerable performance boost when a state-of-the-art detection model is optimized primarily on hybrid sets of real and synthetic data.
  arXiv  Detail & Related papers  (2023-10-25T00:20:37Z)
• Explore the Power of Synthetic Data on Few-shot Object Detection [27.26215175101865]
  Few-shot object detection (FSOD) aims to expand an object detector for novel categories given only a few instances for training. Recent text-to-image generation models have shown promising results in generating high-quality images. This work extensively studies how synthetic images generated from state-of-the-art text-to-image generators benefit FSOD tasks.
  arXiv  Detail & Related papers  (2023-03-23T12:34:52Z)
• Synthetic Data for Object Classification in Industrial Applications [53.180678723280145]
  In object classification, capturing a large number of images per object and in different conditions is not always possible. This work explores the creation of artificial images using a game engine to cope with limited data in the training dataset.
  arXiv  Detail & Related papers  (2022-12-09T11:43:04Z)
• Analysis of Training Object Detection Models with Synthetic Data [0.0]
  This paper attempts to provide a holistic overview of how to use synthetic data for object detection. We analyse aspects of generating the data as well as techniques used to train the models. Experiments are validated on real data and benchmarked to models trained on real data.
  arXiv  Detail & Related papers  (2022-11-29T10:21:16Z)
• Is synthetic data from generative models ready for image recognition? [69.42645602062024]
  We study whether and how synthetic images generated from state-of-the-art text-to-image generation models can be used for image recognition tasks. We showcase the powerfulness and shortcomings of synthetic data from existing generative models, and propose strategies for better applying synthetic data for recognition tasks.
  arXiv  Detail & Related papers  (2022-10-14T06:54:24Z)
• Synthetic Data for Model Selection [2.4499092754102874]
  We show that synthetic data can be beneficial for model selection. We introduce a novel method to calibrate the synthetic error estimation to fit that of the real domain.
  arXiv  Detail & Related papers  (2021-05-03T09:52:03Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.