update readme

# Fun

 No newline at end of file

# Spatial Attention Pyramid Network for Unsupervised Domain Adaptation

## Introduction

This paper was accepted to ECCV 2020.

Unsupervised domain adaptation is critical in various computer vision tasks, such as object detection, instance segmentation, and semantic segmentation, which aims to alleviate performance degradation caused by domain-shift. Most of previous methods rely on a single-mode distribution of source and target domains to align them with adversarial learning, leading to inferior results in various scenarios. To that end, in this paper, we design a new spatial attention pyramid network for unsupervised domain adaptation. Specifically, we first build the spatial pyramid representation to capture context information of objects at different scales. Guided by the task-specific information, we combine the dense global structure representation and local texture patterns at each spatial location effectively using the spatial attention mechanism. In this way, the network is enforced to focus on the discriminative regions with context information for domain adaption. We conduct extensive experiments on various challenging datasets for unsupervised domain adaptation on object detection, instance segmentation, and semantic segmentation, which demonstrates that our method performs favorably against the state-of-the-art methods by a large margin.

![SAPNET](SAPNET.png)

import torch

import numpy as np

import matplotlib.pyplot as plt

from torch import nn

from torch.autograd import Variable

import torch.nn.functional as F

from functools import reduce, partial

from detection.layers import style_pool2d

if __name__ == '__main__':

    a = torch.randn(1, 32, 5, 5)

    y1 = style_pool2d(a, kernel_size=5)

    var, mean = torch.var_mean(a, dim=(2, 3), unbiased=False, keepdim=True)  # (n, c, new_h * new_w)

    std = torch.sqrt(var + 1e-12)

    windows = torch.cat((mean, std), dim=1)

    print(torch.allclose(y1, windows))

    print((y1 == windows))

                writer.add_scalar('lr', optimizer.param_groups[0]['lr'], global_step=global_step)

                writer.add_scalar('LAMBDA', LAMBDA, global_step=global_step)

        if global_step % (500 // max(1, (dist_utils.get_world_size() // 2))) == 0 and test_func is not None:

        if global_step % (2000 // max(1, (dist_utils.get_world_size() // 2))) == 0 and test_func is not None:

            updated = test_func()

            if updated:

                save_func('best.pth', 'mAP: {:.4f}'.format(best_mAP))

                for dataset_name, metric in metrics.items():

                    for k, v in metric.items():

                        metrics_writers[dataset_name].add_scalar('metrics/' + k, v, global_step=global_step)

                        if k == 'mAP' and v > best_mAP:

                        # if k == 'AP50' and v > best_mAP:

                        # if k == 'mAP' and v > best_mAP:

                        if k == 'AP' and v > best_mAP:

                            best_mAP = v

                            updated = True

            model.train()