MGN结构中间层可视化结果怎么解释？

[JR-Wang]

你好，我用Market1501数据集训练mgn_R50-ibn模型，测试的mAP和给出的结果值一致。我从如下b1分支取中间层结果绘制成热图，发现取到b1[0][4].conv3为止，热图的关注位置位于人体上；但是到b1[1][2].conv3，也就是输入到head层之前的最后一个卷积层，用不同测试图像绘制出的热图关注位置变为图像中的固定位置，且不在人体区域上。我把绘制出的热图放在下面，请教一下这种情况要怎么解释？ (b1): Sequential( (0): Sequential( (0): Bottleneck( (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn1): IBN( (IN): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=False) (BN): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) ) (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) (bn2): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn3): BatchNorm(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) (se): Identity() ) (1): Bottleneck( (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn1): IBN( (IN): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=False) (BN): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) ) (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) (bn2): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn3): BatchNorm(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) (se): Identity() ) (2): Bottleneck( (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn1): IBN( (IN): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=False) (BN): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) ) (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) (bn2): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn3): BatchNorm(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) (se): Identity() ) (3): Bottleneck( (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn1): IBN( (IN): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=False) (BN): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) ) (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) (bn2): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn3): BatchNorm(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) (se): Identity() ) (4): Bottleneck( (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn1): IBN( (IN): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=False) (BN): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) ) (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) (bn2): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn3): BatchNorm(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) (se): Identity() ) ) (1): Sequential( (0): Bottleneck( (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn1): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) (bn2): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn3): BatchNorm(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) (se): Identity() (downsample): Sequential( (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False) (1): BatchNorm(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) ) ) (1): Bottleneck( (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn1): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) (bn2): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn3): BatchNorm(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) (se): Identity() ) (2): Bottleneck( (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn1): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) (bn2): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False) (bn3): BatchNorm(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) (se): Identity() ) ) ) b1[0][4].conv3可视化结果 b1[1][2].conv3可视化结果

[L1aoXingyu]

你可以看看 mgn 原始论文的解释，由于 mgn 后两个分支做了分块，也许是这个原因。

[github-actions[bot]]

This issue is stale because it has been open for 30 days with no activity.

[github-actions[bot]]

This issue was closed because it has been inactive for 14 days since being marked as stale.