A3Mark Evaluation Results - Version 1

We compute quality measures with respect to the four ground truth seismic cubes depending on their discrete or continuous amplitude values. The metrics are summarized below, for more precise mathematical details, please refer to our paper [1].

Discrete group:

  • precision: Percentage of true positives of extracted edges/faults in attributes to all extracted edges/faults.
  • recall: Percentage of true positives of extracted edges/faults in attributes to all ground truth edges/faults.
  • rmsDistance: Root mean square weighted by the nonlinear distance (sigmf([1:1:30],[0.5 10])) between edges/faults in attributes and their closest edges/faults in ground truth. Extracted edges/faults using the seismic attributes are treated differently depending on their distance to the closest object in the ground truth.

Continuous group:

  • recall: Percentage of true positives of extracted attributes to ground truth.
  • rms: Root mean square between attributes and ground truth.
  • rmsDiscontinuity: Root mean square between attributes and ground truth without including the edges/faults area. This originates from the fact that most structural attributes (except edge attributes) are undefined at (or very near) edges or discontinuities. This is because derivatives are undefined at discontinuities.

The threshold for a true positive is different for each ground truth category. For each continuous group, the threshold D is 20% of the difference between the maximum and minimum of its ground truth. If the submitted seismic attribute for edge/fault is continuous, not discrete, they will be discretized according to the top 20% rule.

We have computed their varieties, both in 3D and 2D, as well as in different dip regions [0°, 90°] and [0°, 45°]. In the 3D version, the quality measures are computed over the entire seismic cube, whereas in 2D they are first computed on every inline, crossline, and time section of the cube, followed by a mean operation. A caveat of the above metrics is that they do not discriminate between attributes at different dip angle ranges. However, in practice, it is very hard to image structural dips beyond 45 degrees for most types of seismic data. Therefore, more quality metrics (named *_LowDip), which account only for the dip in the 0-45 degrees range, are developed.

Submit and evaluate your own attributes.

Dip_angle
Attribute Name Updated Average Train
N/F
Train
N/R
Train
N/C
Test 1
N/F
Test 1
N/R
Test 1
N/C
Test 2
N/F
Test 2
N/R
Test 2
N/C
SLB_Petrel_Consistent dip_Xing 2016/12/11 0.4984 0.7713 0.0701 0.6701 0.7643 0.0699 0.6631 0.7536 0.0697 0.6532
SLB_Petrel_Local structural dip_Xing 2016/12/11 0.1787 0.1533 0.1768 0.2086 0.1526 0.1759 0.2068 0.1531 0.1756 0.2054
Attribute Name Updated Average Train
N/F
Train
N/R
Train
N/C
Test 1
N/F
Test 1
N/R
Test 1
N/C
Test 2
N/F
Test 2
N/R
Test 2
N/C
SLB_Petrel_Consistent dip_Xing 2016/12/11 19.2649 11.5208 33.3597 12.3112 11.9465 33.2766 12.7974 11.9838 33.3694 12.8191
SLB_Petrel_Local structural dip_Xing 2016/12/11 8.1654 8.0685 8.1661 8.0917 8.1353 8.2305 8.1646 8.1680 8.2646 8.1995
Attribute Name Updated Average Train
N/F
Train
N/R
Train
N/C
Test 1
N/F
Test 1
N/R
Test 1
N/C
Test 2
N/F
Test 2
N/R
Test 2
N/C
SLB_Petrel_Consistent dip_Xing 2016/12/11 19.2291 11.7228 34.3816 12.4161 11.3664 33.6365 12.1685 11.3799 33.8315 12.1583
SLB_Petrel_Local structural dip_Xing 2016/12/11 8.1009 7.9647 8.0667 7.9975 8.0103 8.1092 8.0489 8.1920 8.2898 8.2294
Attribute Name Updated Average Train
N/F
Train
N/R
Train
N/C
Test 1
N/F
Test 1
N/R
Test 1
N/C
Test 2
N/F
Test 2
N/R
Test 2
N/C
SLB_Petrel_Consistent dip_Xing 2016/12/11 15.0059 7.9082 27.3021 9.1703 8.4005 27.2149 9.6049 8.4671 27.3302 9.6546
SLB_Petrel_Local structural dip_Xing 2016/12/11 7.8223 7.6852 7.8067 7.7547 7.7823 7.8930 7.8521 7.8143 7.9267 7.8856
Attribute Name Updated Average Train
N/F
Train
N/R
Train
N/C
Test 1
N/F
Test 1
N/R
Test 1
N/C
Test 2
N/F
Test 2
N/R
Test 2
N/C
SLB_Petrel_Consistent dip_Xing 2016/12/11 14.4338 7.2468 27.5901 8.6683 7.1328 27.2470 8.6313 7.1848 27.5098 8.6929
SLB_Petrel_Local structural dip_Xing 2016/12/11 7.7580 7.6195 7.7479 7.6982 7.6507 7.7759 7.7359 7.7944 7.9196 7.8800
Attribute Name Updated Average Train
N/F
Train
N/R
Train
N/C
Test 1
N/F
Test 1
N/R
Test 1
N/C
Test 2
N/F
Test 2
N/R
Test 2
N/C
SLB_Petrel_Consistent dip_Xing 2016/12/11 0.6382 0.8910 0.2924 0.7596 0.8789 0.2869 0.7474 0.8637 0.2869 0.7371
SLB_Petrel_Local structural dip_Xing 2016/12/11 0.3488 0.3268 0.3512 0.3812 0.3219 0.3458 0.3745 0.3225 0.3445 0.3711
Attribute Name Updated Average Train
N/F
Train
N/R
Train
N/C
Test 1
N/F
Test 1
N/R
Test 1
N/C
Test 2
N/F
Test 2
N/R
Test 2
N/C
SLB_Petrel_Consistent dip_Xing 2016/12/11 4.1535 3.2278 5.1851 3.8309 3.3244 5.2854 3.9072 3.3611 5.3298 3.9297
SLB_Petrel_Local structural dip_Xing 2016/12/11 5.3779 5.3000 5.3425 5.3496 5.3625 5.4014 5.4181 5.3668 5.4164 5.4434
Attribute Name Updated Average Train
N/F
Train
N/R
Train
N/C
Test 1
N/F
Test 1
N/R
Test 1
N/C
Test 2
N/F
Test 2
N/R
Test 2
N/C
SLB_Petrel_Consistent dip_Xing 2016/12/11 3.7730 2.7178 4.9192 3.5741 2.7412 4.9671 3.6054 2.7700 5.0291 3.6333
SLB_Petrel_Local structural dip_Xing 2016/12/11 5.3191 5.2092 5.2541 5.2415 5.2933 5.3373 5.3326 5.3743 5.4187 5.4105
Attribute Name Updated Average Train
N/F
Train
N/R
Train
N/C
Test 1
N/F
Test 1
N/R
Test 1
N/C
Test 2
N/F
Test 2
N/R
Test 2
N/C
SLB_Petrel_Consistent dip_Xing 2016/12/11 4.1068 3.0616 5.4063 3.5947 3.1599 5.5032 3.6600 3.2413 5.6118 3.7228
SLB_Petrel_Local structural dip_Xing 2016/12/11 6.5384 6.8938 6.7134 6.6481 6.9331 6.7538 6.6588 6.0217 6.0927 6.1299
Attribute Name Updated Average Train
N/F
Train
N/R
Train
N/C
Test 1
N/F
Test 1
N/R
Test 1
N/C
Test 2
N/F
Test 2
N/R
Test 2
N/C
SLB_Petrel_Consistent dip_Xing 2016/12/11 3.1330 1.5740 4.9350 2.7325 1.5949 4.9684 2.7512 1.6867 5.1094 2.8445
SLB_Petrel_Local structural dip_Xing 2016/12/11 5.4314 5.3149 5.3654 5.3139 5.3488 5.3989 5.3474 5.5766 5.6214 5.5953
Ground Truth vs Attribute Result (Snapshots from different view angles. Please single click the table cell above.)