text
stringlengths 0
820
|
---|
is 2x2 with stride 2 that outputs a feature map at 2x the in- |
put resolution (28 in Figure 2), followed by a LayerNorm |
and GELU, and then another 2x2 deconvolution layer that |
outputs a feature maps at 2x the previous resolution (56 in |
Figure 2). See the supplementary material for a full architec- |
tural diagram. |
Reconstruction After having been upsampled, the lower |
resolution and higher resolution feature maps are passed into |
Laplacian Blocks (LBs in Figure 2) that reconstruct high |
and low resolution images for the high and low frequency |
reconstruction, respectively. Architecturally, the Laplacian |
Blocks consist of a sequence of three sub-blocks: a Lapla- |
cian Feature Mapping Block, a Laplacian Upsample Block, |
and a Laplacian Pyramid Reconstruction Block. The Feature |
Mapping Block is used to project features within a particular |
layer of the Laplacian Pyramid back to the RGB space. The |
Laplacian Upsample Block represents a learnable upsam- |
ple function that maps latent features from one layer of the |
Laplacian Pyramid to a higher level. Finally, the Laplacian |
Pyramid Reconstruction Block is used to reconstruct infor- |
mation at the different frequencies in RGB space. Following |
super resolution literature [2], an L1 loss is used for high |
frequency output to better reconstruct edges and an L2 loss |
is used for low frequency output to better reconstruct aver- |
age values. The supplementary material has architectural |
diagrams for each block. |
4. Experiments |
We investigate the quality of representations learned from |
Scale-MAE pretraining through a set of experiments that |
explore their robustness to scale as well as their transfer |
performance to additional tasks. First, we present our main |
experiments in Section 4.1 and compare with SatMAE [13], |
a current state-of-the-art MAE for remote sensing imagery, |
Input Image Mask Low Frequency High Frequency ReconstructionFigure 4. Scale-MAE reconstruction. Examples from Functional |
Map of the World are shown. From left to right, an input image |
at 224x224 resolution is shown. Its corresponding mask is visual- |
ized as well. Columns 3 and 4 show the low and high frequency |
produced by the Scale-MAE decoder. The last column is the re- |
construction obtained from summing the low and high frequency |
features together. |
ConvMAE [21], a state-of-the-art multiscale MAE, as well |
as several other approaches detailed throughout. The exact |
implementation of Scale-MAE for the main experiments was |
determined through a set of ablation experiments presented |
in Section 4.2. |
We pretrain a ViT-Large model with Scale-MAE using the |
Functional Map of the World (FMoW) [12] RGB training set, |
which consists of 363.6k images of varying image resolution |
and GSD, for 800 epochs. The initial higher resolution image |
Ihris taken as a random 448px2crop of the input image, and |
the input image Iis then a downsampled 224px2fromIhr. |
The low frequency groundtruth is obtained by downscaling |
Ihrto 14px2and then upscaling to 224px2, while the high |
frequency groundtruth is obtained by downscaling Ihrto |
56px2and then upscaling to 448px2and subtracting this |
image from Ihr. |
Figure 4 shows examples of the masked input, low resolu- |
tion/frequency, high resolution/frequency, and combined re- |
construction of FMoW images during training. The low res- |
olution/frequency images capture color gradients and land- |
scapes, while the residual high resolution/frequency images |
capture object edges, roads, and building outlines. |
4.1. Representation Quality |
We evaluate the quality of representations from |
Scale-MAE by freezing the encoder and performing a non- |
parametric k-nearest-neighbor (kNN) classification with |
eight different remote sensing imagery classification datasets |
0 25% 50% 75% 100%0.50.60.70.80.91.0KNN acc. |
RESISC |
Scale-MAE |
SatMAE |
ConvMAE |
0 25% 50% 75% 100%0.50.60.70.80.91.0 |
Optimal-31 |
0 25% 50% 75% 100%0.50.60.70.80.91.0 |
MLRSNet |
0 25% 50% 75% 100%0.50.60.70.80.91.0 |
CV-BrCT |
0 25% 50% 75% 100% |
Relative GSD0.50.60.70.80.91.0KNN acc. |
WHU-RS19 |
0 25% 50% 75% 100% |
Relative GSD0.50.60.70.80.91.0 |
EuroSAT |
0 25% 50% 75% 100% |
Relative GSD0.50.60.70.80.91.0 |
AiRound |
0 25% 50% 75% 100% |
Relative GSD0.50.60.70.80.91.0 |
UC MercedFigure 5. Learning better representations at all scales. Scale-MAE (blue) features perform better than state-of-the-art. We evaluate kNN |
accuracy on eight datasets with a large variance in GSD. Scale-MAE consistently produces better results at coarser resolutions. In addition |
to using evaluation datasets at different GSDs, to further test the multiscale representations, we create multiple test sets for each dataset |
in which we downsampled the full resolution validation set to coarser GSDs at fixed percentages: XG% |
val, G∈ {12.5,25,50,100}, where |
EuroSat does not include the 12.5% because the images are at a resolution of 64px, our patch size is 16px, and an input image of 8px is too |
small. |
with different GSDs, none of which were encountered dur- |
ing pretraining. The kNN classifier operates by encoding |
all train and validation instances, where each embedded in- |
stance in the validation set computes the cosine distance |