gaussian parameters

[chun]

Hi, I want to know where does the introduction of gaussian parameters for uncertainty calculation?
Thanks.

[xuannianz]

keras-GaussianYOLOv3/model.py

Lines 115 to 121 in b5c321e

 x, y1 = make_last_layers(darknet.output, 512, num_anchors_per_layer * (num_classes + 9)) 
 x = compose(darknet_conv2d_bn_leaky(256, (1, 1)), UpSampling2D(2))(x) 
 x = Concatenate()([x, darknet.layers[152].output]) 
 x, y2 = make_last_layers(x, 256, num_anchors_per_layer * (num_classes + 9)) 
 x = compose(darknet_conv2d_bn_leaky(128, (1, 1)), UpSampling2D(2))(x) 
 x = Concatenate()([x, darknet.layers[92].output]) 
 x, y3 = make_last_layers(x, 128, num_anchors_per_layer * (num_classes + 9)) 

Here you can see there are num_classes + 9 params for each anchor instead of num_classes + 5 in original YOLOv3.
The first 8 params are _{tx ty tw th tx ty tw th}.

[chun]

Thanks.
But I still confused that how an where do you define the gaussian distribution and use its parameters for uncertainty prediction?

[xuannianz]

The gaussian distribution expression is defined in loss function.

[chun]

Sorry, I still cannot localize the gaussian definition, and don't understanding how it work for uncertainty calculation. Could you please tell me in detail?
Thanks.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[chun]

Thanks. If I want to test model flops and FPS, have your code provide these computations? 发送自 Windows 10 版邮件应用 发件人: xuannianz 发送时间: 2019年11月20日 8:54 收件人: xuannianz/keras-GaussianYOLOv3 抄送: ChunmianLin; Author 主题: [SPAM] Re: [xuannianz/keras-GaussianYOLOv3] gaussianparameters (#2) The gaussian distribution expression is defined in loss function. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.