landskape-ai / triplet-attention Goto Github PK

line 293, in parse_model
m_ = nn.Sequential(*(m(*args) for _ in range(n))) if n > 1 else m(*args) # module
TypeError: init() takes from 1 to 2 positional arguments but 3 were given

您好，我在yolov5中想加入这个注意力，然后报这个错误，请问该如何修改呢？

triplet-attention.py

Hi,Run this file and find inconsistent dimensions

I wonder if there is a problem here

torch.cat( (torch.max(x, 1)[0].unsqueeze(1), torch.mean(x, 1).unsqueeze(1)), dim=1 )

it should be torch.cat( (torch.max(x, 1).unsqueeze(1), torch.mean(x, 1).unsqueeze(1)), dim=1 )

right?

我这边使用res2net的话，看到您这边没有对应的res2net+triplet_attentation预训练权重，会有影响吗？

Triplet Attention (k = 7)， what is the parameter ‘k’ means ?

请问根目录下的triplet_attention.py和model文件下的triplet_attention.py哪个效果更好呢

hello my friend!!!
When I download the weight in your readme.md file ,a .rar file was downloaded but I can't open it or unzip it.
the error is :

Thanks for your code!
When I try to add this attention mechanism into the resnet backbone and train a detection network, the error appears as:

File "anaconda3/envs/maskrcnn/lib/python3.7/site-packages/torch/utils/data/_utils/signal_handling.py", line 63, in handler
_error_if_any_worker_fails()
RuntimeError: DataLoader worker (pid 13791) is killed by signal: Aborted.

The batch size and num_workers have been set as 1 and it stills goes wrong.
Could you give some suggestions?

Thank you for sharing your work, the interaction of information between different dimensions is very interesting and this is very helpful for me.

When I tried to reproduce Triplet attention on resnet20,32, I had a little problem, the accuracy of my baseline model as well as the model after applying Triplet attention were lower than the results mentioned in your article. So can you share your pre-training weights and experimental setup?
Thanks a lot!

Triplet Attention (k = 7)， what is the parameter ‘k’ means ?

in triplet_attention.py:

class TripletAttention(nn.Module):
    def __init__(self, gate_channels, reduction_ratio=16, pool_types=['avg', 'max'], no_spatial=False):
        super(TripletAttention, self).__init__()
        self.ChannelGateH = SpatialGate()
        self.ChannelGateW = SpatialGate()
        self.no_spatial = no_spatial
        if not no_spatial:
            self.SpatialGate = SpatialGate()
    
    def forward(self, x):
        x_perm1 = x.permute(0, 2, 1, 3).contiguous()
        x_out1 = self.ChannelGateH(x_perm1)
        x_out11 = x_out1.permute(0, 2, 1, 3).contiguous()
        x_perm2 = x.permute(0, 3, 2, 1).contiguous()
        x_out2 = self.ChannelGateW(x_perm2)
        x_out21 = x_out2.permute(0, 3, 2, 1).contiguous()
        if not self.no_spatial:
            x_out = self.SpatialGate(x)
            x_out = (1 / 3) * (x_out + x_out11 + x_out21)
        else:
            x_out = (1 / 2) * (x_out11 + x_out21)
        return x_out

seems like gate_channels, reduction_ratio, pool_types are never used?

Hi,
For the usage, I didn't find the way to enable to multi-GPU training option. I attached one screenshot.

When the operation order was changed from cw > hc > hw to hw > cw > hc, performance improved in certain models. Calculations are performed independently of each other. Do you know why this is?

thank you.

Thank the author for sharing your work, which is very helpful to me.

I have a little question after studying your work

Is your attention level added after each block? Is there a theoretical difference between this and only adding at the last level?

Hi,

No matter which definitions that I used for triplet-attention definition, I can't load the pertaining weights.

And could you please let me know which the triplet-attention version that matches the Imagenet Pretraining weights?

model = ResidualNet( 'ImageNet', 18, 1000, 'TripletAttention')
state_dict = torch.load("model_best.pth.tar")['state_dict']. #ResNet-18+Triplet Attention(k =3)
model.load_state_dict(state_dict)

Unexpected key(s) in state_dict: "module.conv1.weight", "module.bn1.weight", "module.bn1.bias", "module.bn1.running_mean", "module.bn1.running_var", "module.bn1.num

other double,
k=3 mean that SpatialGate module kernel size=3?

Hi,
I came across the following error. Does anyone has similar experience?
Thanks.

hello, i applied triple_attention on fasterrcnn using mmdetection, the problem is inference speed becomes very slow ， about twice slower，why this happens？ And where is the code you applied trilet_attention on fasterrcnn using mmdetection？i‘d like to learn some. Thanks~

Hi,I wanna ask that how to calculate the AP value of. I can't find it in the code. And I don't know how to calculate the Recall value of.
Thank you very much.

如题，有没有方法直接一键替换SE和CBAM

triplet_attention模块初始化并不是与SE一样是channels参数

# SE attention
class SE(nn.Module):
    def __init__(self, in_channels, channels, se_ratio=12):
        super(SE, self).__init__()
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        self.fc = nn.Sequential(
            nn.Conv2d(in_channels, channels // se_ratio, kernel_size=1, padding=0),
            nn.BatchNorm2d(channels // se_ratio),
            nn.ReLU(inplace=True),
            nn.Conv2d(channels // se_ratio, channels, kernel_size=1, padding=0),
            nn.Sigmoid()
        )

    def forward(self, x):
        y = self.avg_pool(x)
        y = self.fc(y)
        return x * y

1. First of all, thank you very much for the triplet attention. After verification on the ReID task, the effect is really good. The disadvantage is that the running time is slow.
2. Try to use this triplet approach to SE too, the effect is also okay
   3.For this feature map, find the mean along the direction of the h channel, then when you look at a certain channel alone, as shown in the figure, the outermost channel with people, when finding the mean, the person’s characteristics are calculated from head to toe. It feels very unreasonable.
   Can you talk about the inspiration for doing this, or explain it, I am confused about it
   

模型定义的名称和预训练模型的state_dict里的名称不一致导致无法加载。
具体的，原始的定义名称应为：
1、在Class SpatialGate()中，
self.conv = BasicConv(2, 1, kernel_size, stride=1, padding=(kernel_size-1) // 2, relu=False)。
2、在Class TripletAttention()中，
self.cw = SpatialGate()
self.hc = SpatialGate()
self.hw = SpatialGate()
3、在resnet的block定义中，
self.triplet = TripletAttention(planes, 16)