Table 2 Experimental results on the tiny (a), base (b), and large (c) models with the KITTI dataset. Values in bold represent the best results between the baseline and the optimisations applied for that model. The lines highlighted in italic show the best trade-offs between RMSE and inference time considering all models. These points are those values that compose the Pareto frontier.
From: Efficient attention vision transformers for monocular depth estimation on resource-limited hardware
Model | RMSE [m] \(\downarrow\) | \(Abs_{Rel} \downarrow\) | \(\delta _1 \uparrow\) | \(\delta _2 \uparrow\) | \(\delta _3 \uparrow\) | I9X79 [s] \(\downarrow\) | I9X10 [s] \(\downarrow\) | XG38 [s] \(\downarrow\) |
|---|---|---|---|---|---|---|---|---|
(a) | ||||||||
METER | 5.945 | 7.408 | 0.287 | 0.485 | 0.604 | 26.16 | 18.04 | 8.13 |
Meta METER | 6.065 | 8.268 | 0.312 | 0.506 | 0.617 | 20.87 | 17.95 | 15.38 |
Pyra METER | 6.048 | 7.615 | 0.312 | 0.502 | 0.612 | 23.61 | 17.87 | 17.91 |
MoH METER | 6.173 | 9.508 | 0.320 | 0.506 | 0.615 | 31.97 | 20.06 | 18.75 |
PXF | 2.324 | 0.060 | 0.966 | 0.996 | 0.999 | 468.91 | 256.89 | 160.90 |
Meta PXF | 4.231 | 0.125 | 0.837 | 0.956 | 0.987 | 394.75 | 193.13 | 126.90 |
Meta-Base PXF | 4.312 | 0.126 | 0.833 | 0.953 | 0.987 | 420.84 | 209.51 | 136.83 |
Base-Meta PXF | 2.335 | 0.060 | 0.964 | 0.995 | 0.999 | 456.77 | 236.17 | 145.98 |
Pyra PXF | 3.150 | 0.087 | 0.915 | 0.984 | 0.996 | 482.70 | 243.56 | 171.71 |
Pyra-Base PXF | 3.140 | 0.084 | 0.916 | 0.985 | 0.996 | 482.27 | 241.43 | 158.92 |
Base-Pyra PXF | 2.310 | 0.060 | 0.964 | 0.996 | 0.999 | 499.32 | 259.68 | 161.13 |
MoH PXF | 3.516 | 0.095 | 0.895 | 0.976 | 0.994 | 533.67 | 275.82 | 171.57 |
MoH-Base PXF | 3.600 | 0.097 | 0.892 | 0.977 | 0.994 | 525.27 | 278.92 | 168.56 |
Base-MoH PXF | 3.546 | 0.096 | 0.894 | 0.977 | 0.994 | 516.21 | 261.16 | 161.84 |
NeWCRFs | 2.373 | 0.059 | 0.965 | 0.995 | 0.999 | 847.32 | 463.72 | 234.65 |
Meta NeWCRFs | 6.929 | 0.284 | 0.559 | 0.822 | 0.932 | 505.81 | 235.54 | 156.89 |
Meta-Base NeWCRFs | 4.201 | 0.122 | 0.844 | 0.959 | 0.988 | 780.12 | 427.10 | 207.85 |
Base-Meta NeWCRFs | 6.069 | 0.219 | 0.662 | 0.879 | 0.958 | 611.31 | 280.70 | 180.89 |
Pyra NeWCRFs | 5.140 | 0.177 | 0.738 | 0.920 | 0.976 | 667.40 | 306.87 | 205.90 |
Pyra-Base NeWCRFs | 3.193 | 0.088 | 0.914 | 0.983 | 0.996 | 801.98 | 457.53 | 231.76 |
Base-Pyra NeWCRFs | 4.750 | 0.171 | 0.756 | 0.934 | 0.981 | 697.67 | 312.78 | 200.27 |
MoH NeWCRFs | 2.483 | 0.062 | 0.958 | 0.994 | 0.999 | 704.72 | 355.31 | 223.71 |
MoH-Base NeWCRFs | 2.432 | 0.060 | 0.962 | 0.995 | 0.999 | 821.65 | 468.97 | 245.97 |
Base-MoH NeWCRFs | 2.361 | 0.060 | 0.963 | 0.995 | 0.999 | 687.85 | 347.26 | 214.89 |
(b) | ||||||||
METER | 5.794 | 6.625 | 0.302 | 0.504 | 0.618 | 42.51 | 29.66 | 22.16 |
Meta METER | 5.920 | 7.797 | 0.329 | 0.516 | 0.622 | 44.45 | 24.12 | 18.22 |
Pyra METER | 6.052 | 7.010 | 0.319 | 0.498 | 0.605 | 36.50 | 26.22 | 26.65 |
MoH METER | 5.958 | 8.033 | 0.329 | 0.514 | 0.621 | 44.69 | 31.88 | 25.17 |
PXF | 2.205 | 0.055 | 0.972 | 0.997 | 0.999 | 781.45 | 437.31 | 262.77 |
Meta PXF | 4.161 | 0.120 | 0.845 | 0.956 | 0.987 | 567.21 | 300.64 | 194.17 |
Meta-Base PXF | 4.250 | 0.119 | 0.843 | 0.953 | 0.986 | 585.94 | 317.15 | 200.21 |
Base-Meta PXF | 2.195 | 0.054 | 0.972 | 0.997 | 0.999 | 754.56 | 428.33 | 258.40 |
Pyra PXF | 3.243 | 0.086 | 0.914 | 0.983 | 0.996 | 781.64 | 439.35 | 275.37 |
Pyra-Base PXF | 3.254 | 0.087 | 0.915 | 0.983 | 0.995 | 780.42 | 437.67 | 263.11 |
Base-Pyra PXF | 2.192 | 0.055 | 0.972 | 0.997 | 0.999 | 792.97 | 444.23 | 267.44 |
MoH PXF | 3.561 | 0.094 | 0.894 | 0.975 | 0.993 | 867.64 | 514.36 | 290.09 |
MoH-Base PXF | 3.508 | 0.093 | 0.901 | 0.978 | 0.994 | 870.25 | 502.47 | 295.32 |
Base-MoH PXF | 3.562 | 0.093 | 0.894 | 0.976 | 0.993 | 837.29 | 474.92 | 273.32 |
NeWCRFs | 2.185 | 0.054 | 0.972 | 0.999 | 0.997 | 1152.80 | 718.32 | 356.12 |
Meta NeWCRFs | 7.302 | 0.300 | 0.522 | 0.802 | 0.922 | 679.21 | 372.90 | 221.74 |
Meta-Base NeWCRFs | 4.322 | 0.122 | 0.838 | 0.953 | 0.986 | 876.38 | 530.25 | 270.77 |
Base-Meta NeWCRFs | 6.221 | 0.233 | 0.647 | 0.872 | 0.954 | 877.22 | 483.00 | 279.96 |
Pyra NeWCRFs | 5.311 | 0.186 | 0.725 | 0.914 | 0.973 | 1001.02 | 508.04 | 303.65 |
Pyra-Base NeWCRFs | 3.250 | 0.087 | 0.914 | 0.984 | 0.996 | 1093.31 | 644.32 | 339.26 |
Base-Pyra NeWCRFs | 5.062 | 0.177 | 0.744 | 0.926 | 0.977 | 951.53 | 505.20 | 309.29 |
MoH NeWCRFs | 2.272 | 0.057 | 0.969 | 0.996 | 0.999 | 1041.32 | 557.87 | 342.92 |
MoH-Base NeWCRFs | 2.219 | 0.056 | 0.970 | 0.996 | 0.999 | 1115.38 | 672.96 | 352.22 |
Base-MoH NeWCRFs | 2.203 | 0.055 | 0.971 | 0.996 | 0.999 | 955.48 | 526.33 | 316.13 |
(c) | ||||||||
METER | 5.726 | 7.299 | 0.332 | 0.524 | 0.630 | 44.05 | 30.68 | 34.50 |
Meta METER | 5.711 | 7.469 | 0.270 | .493 | 0.625 | 41.78 | 29.97 | 26.27 |
Pyra METER | 5.744 | 7.122 | 0.288 | 0.501 | 0.621 | 51.37 | 29.30 | 38.01 |
MoH METER | 5.714 | 7.470 | 0.275 | 0.498 | 0.628 | 63.78 | 34.45 | 35.24 |
PXF | 2.123 | 0.052 | 0.975 | 0.997 | 0.999 | 1498.72 | 738.41 | 426.68 |
Meta PXF | 3.426 | 0.094 | 0.892 | 0.977 | 0.995 | 959.37 | 504.63 | 302.80 |
Meta-Base PXF | 3.454 | 0.094 | 0.896 | 0.978 | 0.995 | 984.05 | 510.99 | 304.09 |
Base-Meta PXF | 2.108 | 0.052 | 0.976 | 0.997 | 0.999 | 1314.52 | 729.29 | 388.99 |
Pyra PXF | 3.273 | 0.087 | 0.913 | 0.983 | 0.996 | 1359.07 | 802.24 | 430.41 |
Pyra-Base PXF | 3.202 | 0.086 | 0.913 | 0.983 | 0.996 | 1383.07 | 759.66 | 437.13 |
Base-Pyra PXF | 2.111 | 0.052 | 0.975 | 0.997 | 0.999 | 1575.81 | 748.99 | 412.63 |
MoH PXF | 3.427 | 0.089 | 0.906 | 0.980 | 0.995 | 1781.64 | 830.96 | 462.96 |
MoH-Base PXF | 3.399 | 0.089 | 0.909 | 0.980 | 0.995 | 2491.49 | 833.11 | 473.86 |
Base-MoH PXF | 3.311 | 0.086 | 0.913 | 0.982 | 0.995 | 2111.08 | 785.55 | 429.66 |
NeWCRFs | 2.072 | 0.052 | 0.975 | 0.997 | 0.999 | 1863.63 | 969.38 | 500.17 |
Meta NeWCRFs | 6.952 | 0.288 | 0.545 | 0.815 | 0.930 | 1297.96 | 555.62 | 326.76 |
Meta-Base NeWCRFs | 3.481 | 0.096 | 0.895 | 0.979 | 0.995 | 1423.14 | 725.79 | 374.50 |
Base-Meta NeWCRFs | 5.626 | 0.203 | 0.691 | 0.898 | 0.968 | 1490.97 | 796.73 | 433.06 |
Pyra NeWCRFs | 4.830 | 0.164 | 0.762 | 0.932 | 0.979 | 1730.30 | 837.99 | 477.72 |
Pyra-Base NeWCRFs | 3.302 | 0.087 | 0.910 | 0.982 | 0.996 | 1813.54 | 969.33 | 503.52 |
Base-Pyra NeWCRFs | 4.529 | 0.153 | 0.785 | 0.944 | 0.985 | 1575.83 | 801.59 | 460.03 |
MoH NeWCRFs | 2.176 | 0.053 | 0.972 | 0.997 | 0.999 | 1668.14 | 869.06 | 503.80 |
MoH-Base NeWCRFs | 2.123 | 0.052 | 0.974 | 0.997 | 0.999 | 1709.84 | 992.06 | 518.11 |
Base-MoH NeWCRFs | 2.128 | 0.052 | 0.974 | 0.997 | 0.999 | 1616.46 | 855.36 | 473.46 |