Tpatial-Transformer-Networks-pytorch

Refer to daviddao/spatial-transformer-tensorflow(Tensorflow) .
Implementation of Spatial Transformer Networks.

Statement

Do the Experiments on the cluttered MNIST dataset of daviddao.
The accuracy and loss records can be find in cnn.out & stn.out.
The transform img can be find in transform_img/.
py35_pytorch03_version contains the old version code

Environment

python3.6
pytorch 0.4.0

Accuracy

CNN

Testing: epoch[195/200] loss:0.5264 acc:0.9211
Testing: epoch[196/200] loss:0.5185 acc:0.9194
Testing: epoch[197/200] loss:0.5160 acc:0.9158
Testing: epoch[198/200] loss:0.5053 acc:0.9183
Testing: epoch[199/200] loss:0.5057 acc:0.9153

STN

Testing: epoch[195/200] loss:0.0880 acc:0.9762
Testing: epoch[196/200] loss:0.0961 acc:0.9757
Testing: epoch[197/200] loss:0.0893 acc:0.9742
Testing: epoch[198/200] loss:0.1015 acc:0.9740
Testing: epoch[199/200] loss:0.0938 acc:0.9738

Transform Image

(input|transform|input|transform)