Method

multi-dimensional attention for stereo matching [MDA]


Submitted on 21 Nov. 2023 07:51 by
Z jl (Tsinghua University)

Running time:0.32 s
Environment:1 core @ 2.5 Ghz (Python)

Method Description:
Stereo matching is very important fundamental research in computer vision. Cost aggregation is crucial for the final output--disparity map. Whether the costs at different pixels can be fully aggregated directly determines the credibility and accuracy of matching. Previous networks rely on deep stacking of 2D or 3D convolutional layers to do the job. In this paper, we designed a new attention on the cost volume which can achieve global aggregation across and within disparity costs based on feature information. Moreover, to reduce the complexity of the attention mechanism, we adjusted the internal structure of attention, reducing it from the square complexity of the input image to linear. The designed cost aggregation method is named multi-dimensional attention (MDA), which can directly aggregate the global cost volume, enhance the effect of cost aggregation, and reduce the number of 3D convolutional layers.
Parameters:
alpha = 0
Latex Bibtex:

Detailed Results

This page provides detailed results for the method(s) selected. For the first 20 test images, the percentage of erroneous pixels is depicted in the table. We use the error metric described in Object Scene Flow for Autonomous Vehicles (CVPR 2015), which considers a pixel to be correctly estimated if the disparity or flow end-point error is <3px or <5% (for scene flow this criterion needs to be fulfilled for both disparity maps and the flow map). Underneath, the left input image, the estimated results and the error maps are shown (for disp_0/disp_1/flow/scene_flow, respectively). The error map uses the log-color scale described in Object Scene Flow for Autonomous Vehicles (CVPR 2015), depicting correct estimates (<3px or <5% error) in blue and wrong estimates in red color tones. Dark regions in the error images denote the occluded pixels which fall outside the image boundaries. The false color maps of the results are scaled to the largest ground truth disparity values / flow magnitudes.

Test Set Average

Error D1-bg D1-fg D1-all
All / All 1.37 2.64 1.58
All / Est 1.37 2.64 1.58
Noc / All 1.26 2.58 1.48
Noc / Est 1.26 2.58 1.48
This table as LaTeX

Test Image 0

Error D1-bg D1-fg D1-all
All / All 1.90 1.08 1.79
All / Est 1.90 1.08 1.79
Noc / All 1.90 1.08 1.78
Noc / Est 1.90 1.08 1.78
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 1

Error D1-bg D1-fg D1-all
All / All 1.68 2.86 1.81
All / Est 1.68 2.86 1.81
Noc / All 1.60 2.86 1.74
Noc / Est 1.60 2.86 1.74
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 2

Error D1-bg D1-fg D1-all
All / All 1.79 6.92 2.04
All / Est 1.79 6.92 2.04
Noc / All 1.70 6.92 1.96
Noc / Est 1.70 6.92 1.96
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 3

Error D1-bg D1-fg D1-all
All / All 1.59 0.38 1.48
All / Est 1.59 0.38 1.48
Noc / All 1.59 0.38 1.48
Noc / Est 1.59 0.38 1.48
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 4

Error D1-bg D1-fg D1-all
All / All 0.47 0.33 0.45
All / Est 0.47 0.33 0.45
Noc / All 0.45 0.33 0.43
Noc / Est 0.45 0.33 0.43
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 5

Error D1-bg D1-fg D1-all
All / All 1.81 1.90 1.82
All / Est 1.81 1.90 1.82
Noc / All 1.72 1.90 1.74
Noc / Est 1.72 1.90 1.74
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 6

Error D1-bg D1-fg D1-all
All / All 2.02 1.54 1.97
All / Est 2.02 1.54 1.97
Noc / All 2.07 1.54 2.01
Noc / Est 2.07 1.54 2.01
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 7

Error D1-bg D1-fg D1-all
All / All 0.22 2.43 0.65
All / Est 0.22 2.43 0.65
Noc / All 0.22 2.43 0.66
Noc / Est 0.22 2.43 0.66
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 8

Error D1-bg D1-fg D1-all
All / All 0.25 1.28 0.44
All / Est 0.25 1.28 0.44
Noc / All 0.24 1.28 0.44
Noc / Est 0.24 1.28 0.44
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 9

Error D1-bg D1-fg D1-all
All / All 0.29 1.37 0.57
All / Est 0.29 1.37 0.57
Noc / All 0.29 1.44 0.58
Noc / Est 0.29 1.44 0.58
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 10

Error D1-bg D1-fg D1-all
All / All 1.02 2.42 1.34
All / Est 1.02 2.42 1.34
Noc / All 1.03 2.42 1.35
Noc / Est 1.03 2.42 1.35
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 11

Error D1-bg D1-fg D1-all
All / All 0.80 0.40 0.73
All / Est 0.80 0.40 0.73
Noc / All 0.80 0.40 0.73
Noc / Est 0.80 0.40 0.73
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 12

Error D1-bg D1-fg D1-all
All / All 0.69 0.75 0.70
All / Est 0.69 0.75 0.70
Noc / All 0.53 0.75 0.54
Noc / Est 0.53 0.75 0.54
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 13

Error D1-bg D1-fg D1-all
All / All 0.61 0.37 0.58
All / Est 0.61 0.37 0.58
Noc / All 0.55 0.37 0.53
Noc / Est 0.55 0.37 0.53
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 14

Error D1-bg D1-fg D1-all
All / All 1.29 0.11 1.26
All / Est 1.29 0.11 1.26
Noc / All 1.16 0.11 1.14
Noc / Est 1.16 0.11 1.14
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 15

Error D1-bg D1-fg D1-all
All / All 2.34 0.54 2.18
All / Est 2.34 0.54 2.18
Noc / All 2.39 0.54 2.22
Noc / Est 2.39 0.54 2.22
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 16

Error D1-bg D1-fg D1-all
All / All 3.41 0.24 2.94
All / Est 3.41 0.24 2.94
Noc / All 3.27 0.24 2.82
Noc / Est 3.27 0.24 2.82
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 17

Error D1-bg D1-fg D1-all
All / All 0.78 0.01 0.70
All / Est 0.78 0.01 0.70
Noc / All 0.79 0.01 0.71
Noc / Est 0.79 0.01 0.71
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 18

Error D1-bg D1-fg D1-all
All / All 4.29 1.46 2.95
All / Est 4.29 1.46 2.95
Noc / All 4.23 1.46 2.90
Noc / Est 4.23 1.46 2.90
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 19

Error D1-bg D1-fg D1-all
All / All 0.63 0.41 0.61
All / Est 0.63 0.41 0.61
Noc / All 0.64 0.41 0.61
Noc / Est 0.64 0.41 0.61
This table as LaTeX

Input Image

D1 Result

D1 Error




eXTReMe Tracker