Task2Sim : Towards Effective Pre-training and Transfer from Synthetic Data

11/30/2021
by   Samarth Mishra, et al.
2

Pre-training models on Imagenet or other massive datasets of real images has led to major advances in computer vision, albeit accompanied with shortcomings related to curation cost, privacy, usage rights, and ethical issues. In this paper, for the first time, we study the transferability of pre-trained models based on synthetic data generated by graphics simulators to downstream tasks from very different domains. In using such synthetic data for pre-training, we find that downstream performance on different tasks are favored by different configurations of simulation parameters (e.g. lighting, object pose, backgrounds, etc.), and that there is no one-size-fits-all solution. It is thus better to tailor synthetic pre-training data to a specific downstream task, for best performance. We introduce Task2Sim, a unified model mapping downstream task representations to optimal simulation parameters to generate synthetic pre-training data for them. Task2Sim learns this mapping by training to find the set of best parameters on a set of "seen" tasks. Once trained, it can then be used to predict best simulation parameters for novel "unseen" tasks in one shot, without requiring additional training. Given a budget in number of images per class, our extensive experiments with 20 diverse downstream tasks show Task2Sim's task-adaptive pre-training data results in significantly better downstream performance than non-adaptively choosing simulation parameters on both seen and unseen tasks. It is even competitive with pre-training on real images from Imagenet.

READ FULL TEXT

page 1

page 3

page 7

page 8

page 13

page 16

research
06/21/2022

Insights into Pre-training via Simpler Synthetic Tasks

Pre-training produces representations that are effective for a wide rang...
research
03/10/2022

Knowledge Distillation as Efficient Pre-training: Faster Convergence, Higher Data-efficiency, and Better Transferability

Large-scale pre-training has been proven to be crucial for various compu...
research
07/11/2022

PSP-HDRI+: A Synthetic Dataset Generator for Pre-Training of Human-Centric Computer Vision Models

We introduce a new synthetic data generator PSP-HDRI+ that proves to be ...
research
04/28/2022

Resource-efficient domain adaptive pre-training for medical images

The deep learning-based analysis of medical images suffers from data sca...
research
03/16/2023

A New Benchmark: On the Utility of Synthetic Data with Blender for Bare Supervised Learning and Downstream Domain Adaptation

Deep learning in computer vision has achieved great success with the pri...
research
06/06/2021

A Pre-training Oracle for Predicting Distances in Social Networks

In this paper, we propose a novel method to make distance predictions in...
research
07/27/2023

Pre-training Vision Transformers with Very Limited Synthesized Images

Formula-driven supervised learning (FDSL) is a pre-training method that ...

Please sign up or login with your details

Forgot password? Click here to reset