Table 4 Number of network parameters and memory footprint relative to the tiny (a), base (b), and large (c) size. Values in bold represent the best values between the baseline and the optimisations applied for that model.
From: Efficient attention vision transformers for monocular depth estimation on resource-limited hardware
Model | Parameters [M] \(\downarrow\) | Memory [MB] \(\downarrow\) |
|---|---|---|
(a) | ||
METER | 0.71 | 2.72 |
Meta METER | 0.68 | 2.60 |
Pyra METER | 1.08 | 4.11 |
MoH METER | 0.71 | 2.73 |
PXF | 88.91 | 339.15 |
Meta PXF | 76.08 | 290.22 |
Meta-Base PXF | 80.26 | 306.18 |
Base-Meta PXF | 84.72 | 323.19 |
Pyra PXF | 103.12 | 393.38 |
Pyra-Base PXF | 97.55 | 372.11 |
Base-Pyra PXF | 94.48 | 360.42 |
MoH PXF | 89.03 | 339.61 |
MoH-Base PXF | 88.98 | 339.44 |
Base-MoH PXF | 88.95 | 339.33 |
NeWCRFs | 88.46 | 337.43 |
Meta NeWCRFs | 74.23 | 283.18 |
Meta-Base NeWCRFs | 79.81 | 304.46 |
Base-Meta NeWCRFs | 82.88 | 316.15 |
Pyra NeWCRFs | 111.04 | 423.57 |
Pyra-Base NeWCRFs | 97.09 | 370.39 |
Base-Pyra NeWCRFs | 102.40 | 390.61 |
MoH NeWCRFs | 91.41 | 348.71 |
MoH-Base NeWCRFs | 88.53 | 337.72 |
Base-MoH NeWCRFs | 91.34 | 348.43 |
(b) | ||
METER | 1.45 | 5.53 |
Meta METER | 1.40 | 5.35 |
Pyra METER | 2.29 | 8.74 |
MoH METER | 1.45 | 5.54 |
PXF | 140.43 | 535.71 |
Meta PXF | 108.28 | 413.06 |
Meta-Base PXF | 112.47 | 429.02 |
Base-Meta PXF | 136.25 | 519.75 |
Pyra PXF | 173.96 | 663.62 |
Pyra-Base PXF | 168.39 | 642.34 |
Base-Pyra PXF | 146.01 | 556.98 |
MoH PXF | 140.72 | 536.80 |
MoH-Base PXF | 140.67 | 536.62 |
Base-MoH PXF | 140.48 | 535.88 |
NeWCRFs | 139.98 | 533.99 |
Meta NeWCRFs | 106.44 | 406.02 |
Meta-Base NeWCRFs | 112.01 | 427.30 |
Base-Meta NeWCRFs | 134.40 | 512.71 |
Pyra NeWCRFs | 181.88 | 693.81 |
Pyra-Base NeWCRFs | 167.94 | 640.62 |
Base-Pyra NeWCRFs | 153.92 | 587.17 |
MoH NeWCRFs | 143.10 | 545.90 |
MoH-Base NeWCRFs | 140.22 | 534.90 |
Base-MoH NeWCRFs | 142.86 | 544.98 |
(c) | ||
METER | 3.30 | 12.57 |
Meta METER | 3.22 | 12.29 |
Pyra METER | 5.53 | 21.08 |
MoH METER | 3.30 | 12.58 |
PXF | 270.90 | 1033.39 |
Meta PXF | 203.82 | 777.53 |
Meta-Base PXF | 208.01 | 793.49 |
Base-Meta PXF | 266.71 | 1017.43 |
Pyra PXF | 339.34 | 1294.49 |
Pyra-Base PXF | 333.77 | 1273.22 |
Base-Pyra PXF | 276.47 | 1054.66 |
MoH PXF | 271.47 | 1035.56 |
MoH-Base PXF | 271.42 | 1035.39 |
Base-MoH PXF | 270.94 | 1033.56 |
NeWCRFs | 270.44 | 1031.67 |
Meta NeWCRFs | 201.98 | 770.49 |
Meta-Base NeWCRFs | 207.56 | 791.76 |
Base-Meta NeWCRFs | 264.87 | 1010.39 |
Pyra NeWCRFs | 347.26 | 1324.68 |
Pyra-Base NeWCRFs | 333.32 | 1271.50 |
Base-Pyra NeWCRFs | 284.39 | 1084.85 |
MoH NeWCRFs | 273.85 | 1044.66 |
MoH-Base NeWCRFs | 270.97 | 1033.67 |
Base-MoH NeWCRFs | 273.33 | 1042.66 |
