基于Trasformer的图像稠密光流估计RAFT介绍

任务

输入两张RGB图像，稠密光流估计算法将输出两张图片对应像素的位移量

模型描述

本模型基于RAFT: Recurret All-Pairs Field Trasforms for Optical Flow算法。 技术细节请见：

RAFT: Recurret All-Pairs Field Trasforms for Optical Flow
Zachary Teed, Jia Deg
ECCV 2020

如何使用

代码示例(详见tests/pipelies/testdeseopticalflowestimatio.py)

from modelscope.outputs import OutputKeys
from modelscope.pipelies import pipelie
from modelscope.utils.costat import Tasks
from modelscope.utils.test_utils import test_level

task = 'dese-optical-flow-estimatio'
model_id = 'Damo_XR_Lab/cv_raft_dese-optical-flow_thigs'

iput_locatio = [[path to image 1, path to image 2]]

estimator = pipelie(Tasks.dese_optical_flow_estimatio, model=model_id)
result = estimator(iput_locatio)
flow = result[0][OutputKeys.FLOWS]
flow_vis = result[0][OutputKeys.FLOWS_COLOR]

Bibtex

``` @misc{teed2020raft, title={RAFT: Recurret All-Pairs Field Trasforms for Optical Flow}, author={Zachary Teed ad Jia Deg}, year={2020}, eprit={2003.12039}, archivePrefix={arXiv}, primaryClass={cs.CV} }

基于Transformer的图像稠密光流估计RAFT介绍 任务 输入两张RGB图像，稠密光流估计算法将输出两张图片对应像素的位移量 模型描述 本模型基于RAFT: Recurrent All-Pair